INGEGNERIA ELETTROTECNICA

Transcription

1 Alma Mater Studorum Unverstà d Bologna DOTTORATO DI RICERCA INGEGNERIA ELETTROTECNICA Cclo XX Settore scentfco dscplnare d afferenza: ING INF/07 MISURE ELETTRICHE E ELETTRONICHE DEVELOPMENT AND CHARACTERIZATION OF A DISTRIBUTED MEASUREMENT SYSTEM FOR THE EVALUATION OF VOLTAGE QUALITY IN ELECTRIC POWER NETWORKS Presentata da: ELISA SCALA Coordnatore Dottorato PROF. FRANCESCO NEGRINI Relatore PROF. LORENZO PERETTO Esame fnale anno 2008

2

3 1. Introducton Power qualty n Electrcal Systems Introducton Power Qualty Degradng Phenomena Steady state Voltage Characterstcs Transents Harmonc Dstorton Short Duraton Voltage Varatons Power Qualty Evaluaton Research and Standardzaton Actvty Basc Defntons of Voltage Parameters Overvew of Power Qualty Indces Standard Measurement Methods of PQ Parameters Non conventonal Parameters for PQ measurement European Scenaro: Standard, Gudes and PQ level Relablty Regulaton Voltage Qualty Regulaton Montorng the Voltage Qualty wthn EU Mnmum Standards Incentve Schemes..42 References Electromagnetc Transents n Power Dstrbuton Networks Transents Lghtnng Drect Flashes to Overhead Lnes Induced Overvoltages on Overhead Lnes Overvoltages caused by couplng wth other systems Lghtnng surges transferred from MV systems Surge magntude and propagaton n MV systems Surge transfer to the Low Voltage System Transmsson Lnes Travellng Waves I -

4 Modal Transformaton Tme Frequency Representaton of Sgnals The Wavelet Transform The Dscrete tme Wavelet Transform Flter Banks Wavelets and Flter Banks Fault Locaton Methods for Dstrbuton Networks Sgnal Analyss of Fundamental Frequency Components Sgnal Analyss of Hgh Frequency Components Artfcal Neural Network Technques Dstrbuted Measurement System detectng Transent Dsturbances and Method to locate the Transent Source Procedure to locate the Source of a transent Fault Locaton Method ntegratng the Dstrbuted Measurement System and Wavelet Analyss 93 References Analyss of the performance featured by the fault locaton method based on dstrbuted smultaneous measurements - smulatons of dstrbuton networks n EMTP-RV envronment Implementaton of an IEEE Std. test network affected by bolt faults Expermental Results Lne to lne short crcut Lne to ground short crcut Inserton of capactor bank Implementaton of a dstrbuton network operatng n actual condtons varaton of fault parameters Improvements n the Procedure Test condtons: non deal faults and network elements Models of Power Dstrbuton and Measurement Systems Smulaton Results Drect Lghtnng Phase to ground short crcut Transposed lnes Phase to ground short crcut Untransposed lnes Phase to phase short crcut Untransposed lnes Varatons n the Network Topology II -

5 4.3. Implementaton of a radal dstrbuton network to test the verson of the fault locaton procedure ntegrated wth DWT analyss..129 References The dstrbuted measurement system for transents detecton The Voltage Transducer The Event Detecton Block The GPS based devce What s GPS? The GPS staton The DAQ board The GSM protocol The GPRS protocol Metrologc characterzaton of the measurement system - The Supplement 1 to the G.U.M Man stages of uncertanty evaluaton Propagaton of Dstrbutons Monte Carlo Approach to the propagaton Evaluaton of the combned uncertanty on the poston of the transent source Metrologcal characterzaton of the voltage transducer Metrologcal Characterzaton of the Event Detecton Block Metrologcal Characterzaton of the GPS staton Monte Carlo Method to evaluate the combned uncertanty on the Fault Locaton Expermental Results 163 5A. Appendx: Uncertanty Contrbuton of the Analog Condtonng Block n DSP- based Instruments A.1. 5A.2. 5A.3. 5A.4. 5A.5. The calbraton Method.166 Characterzaton Indces..167 Applcaton Example.170 An Equpment for Voltage Transducers Calbraton 173 Uncertanty sources n the Equpment 176 References Desgn and Characterzaton of an Electrc feld based Medum Voltage Transducer III -

6 6.1. Voltage Transducers Electrc feld strength meters Free body meter Ground reference Type meter Electro-optc meter Calbraton of Electrc feld strength meters Man sources of measurement uncertanty The Voltage Transducer Operatng prncple prototype realzaton Expermental setup voltage error phase error bandwdth mmunty to other electrc feld contrbutons 207 References Conclusons IV -

7 1. INTRODUCTION 1. Introducton The research actvty carred out durng the PhD course n Electrcal Engneerng belongs to the branch of electrc and electronc measurements. The man subject of the present thess s a dstrbuted measurement system to be nstalled n Medum Voltage power networks, as well as the method developed to analyze data acqured by the measurement system tself and to montor power qualty. In chapter 2 the ncreasng nterest towards power qualty n electrcal systems s llustrated, by reportng the nternatonal research actvty nherent to the problem and the relevant standards and gudelnes emtted. The aspect of the qualty of voltage provded by utltes and nfluenced by customers n the varous ponts of a network came out only n recent years, n partcular as a consequence of the energy market lberalzaton. Usually, the concept of qualty of the delvered energy has been assocated mostly to ts contnuty. Hence the relablty was the man characterstc to be ensured for power systems. Nowadays, the number and duraton of nterruptons are the qualty ndcators commonly perceved by most customers; for ths reason, a short secton s dedcated also to network relablty and ts regulaton. In ths contest t should be noted that although the measurement system developed durng the research actvty belongs to the feld of power qualty evaluaton systems, the nformaton regstered n real tme by ts remote statons can be used to mprove the system relablty too. Gven the vast scenaro of power qualty degradng phenomena that usually can occur n dstrbuton networks, the study has been focused on electromagnetc transents affectng lne voltages. The outcome of such a study has been the desgn and realzaton of a dstrbuted measurement system whch contnuously montor the phase sgnals n dfferent ponts of a network, detect the occurrence of transents superposed to the fundamental steady state component and regster the tme of occurrence of such events. The data set s fnally used to locate the source of the transent dsturbance propagatng along the network lnes. Most of the oscllatory transents affectng lne voltages are due to faults occurrng n any pont of the dstrbuton system and have to be seen before protecton equpment nterventon. An mportant concluson s that the method can mprove the montored network relablty, snce the knowledge of the locaton of a fault allows the energy manager to reduce as much as possble both the area of the network to be dsconnected for protecton purposes and the tme spent by techncal staff to recover the abnormal condton and/or the damage. The part of the thess presentng the results of such a study and actvty s structured as follows: chapter 3 deals wth the propagaton of electromagnetc transents n power - 1 -

8 1. INTRODUCTION systems by defnng characterstcs and causes of the phenomena and brefly reportng the theory and approaches used to study transents propagaton. Then the state of the art concernng methods to detect and locate faults n dstrbuton networks s presented. Fnally the attenton s pad on the partcular technque adopted for the same purpose durng the thess, and the methods developed on the bass of such approach. Chapter 4 reports the confguraton of the dstrbuton networks on whch the fault locaton method has been appled by means of smulatons as well as the results obtaned case by case. In ths way the performance featured by the locaton procedure frstly n deal then n realstc operatng condtons are tested. In chapter 5 the measurement system desgned to mplement the transents detecton and fault locaton method s presented. The hardware belongng to the measurement chan of every acquston channel n remote statons s descrbed. Then, the global measurement system s characterzed by consderng the non deal aspects of each devce that can concur to the fnal combned uncertanty on the estmated poston of the fault n the network under test. Fnally, such parameter s computed accordng to the Gude to the Expresson of Uncertanty n Measurements, by means of a numerc procedure. In the last chapter a devce s descrbed that has been desgned and realzed durng the PhD actvty amng at substtutng the commercal capactve voltage dvder belongng to the condtonng block of the measurement chan. Such a study has been carred out amng at provdng an alternatve to the used transducer that could feature equvalent performance and lower cost. In ths way, the economcal mpact of the nvestment assocated to the whole measurement system would be sgnfcantly reduced, makng the method applcaton much more feasble

9 2. POWER QUALITY IN ELECTRICAL SYSTEMS 2. Power qualty n electrcal systems 2.1. Introducton Electrcal energy s a product and, lke any other product, should satsfy the proper qualty requrements. If electrcal equpment s to operate correctly, t requres electrcal energy to be suppled at a voltage that s wthn a specfed range around the rated value. A sgnfcant part of the equpment n use today, especally electronc and computer devces, requres good power qualty (PQ). However, the same equpment often causes dstorton of the voltage supply n the nstallaton, because of ts non-lnear characterstcs,.e. t draws a non-snusodal current wth a snusodal supply voltage. Thus, mantanng satsfactory PQ s a jont responsblty for the suppler and the electrcty user. All customers have to make sure that they obtan an electrcty supply of satsfactory qualty to avod the hgh cost of equpment falures; electrcal equpment must also be capable of functonng as requred when small dsturbances occur. Customers are granted to be safe only f the lmts wthn whch power qualty may vary can be specfed. Accordng to the techncal context, such lmts should be fxed by standards, by the natonal regulator, by the customer by means of a power qualty contract, by the manufacturer n a devce manual or by the grd operator n a gudelne. The defned lmts must be meanngful, consstent and easy to compare to actual power qualty levels. All these dfferent lmts have to be accomplshed, on the one hand to prevent devces or nstallatons from malfunctonng, on the other hand for clear communcaton about the qualty of supply that s provded or demanded. The startng pont for the defnton of the supply-voltage qualty s the set of lmts defned by the Natonal Regulator and to be met at the pont of common couplng wth the customer. At the moment, there s no standard for the current qualty relevant to that pont, n fact the man document dealng wth requrements concernng the suppler s sde s standard EN [1], whch characterzes voltage parameters. Ths s a European standard on electrcal energy n publc dstrbuton systems, completed n some regons or countres by other supplemental standards. Accordng to [1] the suppler s the party who provdes electrcty va a publc dstrbuton system, and the user or customer s the purchaser of electrcty from a suppler. The user s enttled to receve a sutable qualty of power from the suppler. In practce, the level of PQ s a compromse between user and suppler. Where the avalable PQ s not suffcent for the user s needs, PQ mprovement measures are needed and a cost-beneft analyss should be carred out. However, the cost of poor PQ usually exceeds the cost of measures requred for mprovement - t s estmated that losses caused by power qualty degradaton cost EU ndustry and commerce about 10 bllon per annum

10 2. POWER QUALITY IN ELECTRICAL SYSTEMS The electrc system, characterzed n the past by an hgh level of vertcal ntegraton, s the subject of a recent lberalzaton process whch separates the energy producton, the management of the transmsson network and the power dstrbuton among a pluralty of subjects. Besde the classc boundary between the dstrbutor and the user (n hgh, medum or low voltage systems), all the other new boundares must be taken nto account. Ths apples n partcular to voltage qualty, whch has to be defned, at least for the man parameters, not only at supply ponts to the end users, but also at the other boundares. In facts, the fnal voltage qualty depends on the operatng condtons at all levels of the whole process. However, electrcal energy s a very specfc product. The possblty for storng electrcty n any sgnfcant quantty s very lmted, so t s consumed at the same nstant t s generated. Measurement and evaluaton of the qualty of the suppled power has to be made at the nstant of ts consumpton. The measurement of PQ s complex, snce the suppler and user, whose senstve electrcal equpment s also a source of dsturbances, have dfferent perspectves. The nteracton between voltage and current makes t hard to separate the customer as recevng and the network company as supplyng a certan level of PQ. The voltage qualty (for whch the network s often consdered responsble) and current qualty (for whch the customer s often consdered responsble) are affectng each other by mutual nteracton. The effects of nsuffcent PQ are normally expressed n terms of emsson, mmunty and compatblty. The emsson s defned as the causal dsturbance, such as the offset of a voltage from ts nomnal value. The mmunty s the degree at whch the equpment wll be able to functon as planned n spte of the emsson. The compatblty level s the level at whch the rsk of the equpment malfunctonng s suffcently low. On the user s sde, t s the qualty of power avalable to the user s equpment that s mportant. Correct equpment operaton requres the level of electromagnetc nfluence on equpment to be mantaned below certan lmts. Equpment s nfluenced by dsturbances on the supply and by other equpment n the nstallaton, as well as tself nfluencng the supply. These problems are summarzed n the EN seres [7-15] of EMC standards, n whch lmts of conducted dsturbances are characterzed

11 2. POWER QUALITY IN ELECTRICAL SYSTEMS 2.2. Power Qualty degradng phenomena Power qualty varatons fall nto two basc categores: 1. Dsturbances. Dsturbances are measured by trggerng on an abnormalty n the voltage or the current. Transent voltages may be detected when the peak magntude exceeds a specfed threshold. RMS voltage varatons (e.g. sags or nterruptons) may be detected when the RMS varaton exceeds a specfed level. 2. Steady State Varatons. These nclude normal RMS voltage varatons and harmonc dstorton. These varatons must be measured by samplng the voltage and/or current over tme. The nformaton s best presented as a trend of the quantty (e.g. voltage dstorton) over tme and then analyzed usng statstcal methods (e.g. average dstorton level, 95% probablty of not beng exceeded). In the past, measurement equpment has been desgned to handle ether the dsturbances (e.g. dsturbance analyzers) or steady state varatons (e.g. voltage recorders, harmoncs montors). Wth advances n processng capablty, new nstruments have become avalable that can characterze the full range of power qualty varatons. The new challenge nvolves characterzng all the data n a convenent form so that t can be used to help dentfy and solve problems Steady State Voltage Characterstcs There s no such thng as steady state on the power system. Loads are contnually changng and the power system s contnually adjustng to these changes. All of these changes and adjustments result n voltage varatons that are referred to as long duraton voltage varatons. These can be undervoltages or overvoltages, dependng on the specfc crcut condtons. Characterstcs of the steady state voltage are best expressed wth long duraton profles and statstcs. Important characterstcs nclude the voltage ampltude and unbalance. Harmonc dstorton s also a characterstc of the steady state voltage but ths characterstc s treated separately because t does not nvolve varatons n the fundamental frequency component of the voltage. Most end use equpment s not very senstve to these voltage varatons, as long as they are wthn reasonable lmts Transents The term transents s normally used to refer to fast changes n the system voltage or current. Transents are dsturbances, rather than steady state varatons such as harmonc dstorton or voltage unbalance. Dsturbances can be measured by trggerng on the abnormalty nvolved. For transents, t could be the peak magntude, the rate of rse, or just the change n the waveform from one cycle to the next. Transents can be dvded nto two sub-categores, mpulsve transents and oscllatory transents, dependng on - 5 -

12 2. POWER QUALITY IN ELECTRICAL SYSTEMS ther characterstcs. Transents are normally characterzed by the actual waveform, although summary descrptors can also be developed (peak magntude, prmary frequency, rate-of rse, etc.). Fgure 2-1 gves a capactor swtchng transent waveform. Ths s one of the most mportant transents that s ntated on the utlty supply system and can affect the operaton of end user equpment. Transent problems are solved by controllng the transent at the source, changng the characterstcs of the system affectng the transent or by protectng equpment so that t s not mpacted. For nstance, capactor swtchng transents can be controlled at the source by closng the breaker contacts close to a voltage zero crossng. Magnfcaton of the transent can be avoded by not usng low voltage capactors wthn the end user facltes. The actual equpment can be protected wth flters or surge arresters. Fgure 2-1. Dsturbance due to a Capactor swtchng Harmonc Dstorton Harmonc dstorton of the voltage and current results from the operaton of nonlnear loads and devces on the power system. The nonlnear loads that cause harmoncs can often be represented as current sources of harmoncs. The system voltage appears stff to ndvdual loads and the loads draw dstorted current waveforms. Harmonc voltage dstorton results from the nteracton of these harmonc currents wth the system mpedance. The harmonc standard [4], has proposed two crtera for controllng harmonc levels on the power system. In the frst case the end users must lmt the harmonc currents njected onto the power system. In the second soluton the power suppler wll control the harmonc voltage dstorton by makng sure system resonant condtons do not cause excessve magnfcaton of the harmonc levels. Harmonc dstorton levels can be characterzed by the complete harmonc spectrum wth magntudes and phase angles of each ndvdual harmonc component. It s also common to use a sngle quantty, the Total Harmonc Dstorton, as a measure of the magntude of harmonc dstorton. For currents, the dstorton values must be referred to a constant base (e.g. the rated load current or demand current) rather than the fundamental component. Ths provdes a constant reference whle the fundamental can vary over a wde range. Harmonc dstorton s a - 6 -

13 2. POWER QUALITY IN ELECTRICAL SYSTEMS characterstc of the steady state voltage and current. It s not a dsturbance. Therefore, characterzng harmonc dstorton levels s accomplshed wth profles of the harmonc dstorton over tme (e.g. 24 hours) and statstcs. Fgure 2-2 llustrates some example current waveforms for dfferent types of nonlnear loads [4]. Type of load Typcal waveform Current dstorton Sngle phase power supply 80% (hgh thrd) Semconverter Hgh 2nd, 3rd, 4th at partal loads 6 pulse converter, capactve smoothng no seres nductance 80% 6 pulse converter, capactve smoothng wth seres nductance > 3%, or DC drve 40% 6 pulse converter, wth large nductor for current smoothng 28% 12 pulse converter 15% AC voltage regulator Depends on frng angle Fgure 2-2. Current waveforms for nonlnear loads Short Duraton Voltage Varatons Short duraton voltage varatons nclude varatons n the fundamental frequency voltage that last less than one mnute. These varatons are best characterzed by plots of the RMS voltage vs. tme but t s often suffcent to descrbe them by a voltage magntude and a duraton that the voltage s outsde of specfed thresholds. It s usually not necessary to have detaled waveform plots snce the RMS voltage magntude s of - 7 -

14 2. POWER QUALITY IN ELECTRICAL SYSTEMS prmary nterest. The voltage varatons can be a momentary low voltage (voltage sag), hgh voltage (voltage swell), or loss of voltage (nterrupton). Interruptons are the most severe n terms of ther mpacts on end users but voltage sags can be more mportant because they may occur much more frequently. A fault condton can cause a momentary voltage sag over a wde porton of the system even though no end users may experence an nterrupton. Ths s true for most transmsson faults. Many end users have equpment that may be senstve to these knds of varatons. Solvng ths problem on the utlty system may be very expensve so manufacturers are developng rde through technologes wth energy storage to handle these voltage varatons on the end user sde. A voltage dp s specfed n terms of duraton and retaned voltage, usually expressed as the percentage of nomnal RMS voltage remanng at the lowest pont durng the dp. A voltage dp means that the requred energy s not beng delvered to the load and ths can have serous consequences dependng on the type of load nvolved. Voltage sags - longer-term reductons n voltage are usually caused by a delberate reducton of voltage by the suppler to reduce the load at tmes of maxmum demand or by an unusually weak supply n relaton to the load. Motor drves, ncludng varable speed drves, are partcularly susceptble because the load stll requres energy that s no longer avalable except from the nerta of the drve. In processes where several drves are nvolved, ndvdual motor control unts may sense the loss of voltage and shut down the drve at a dfferent voltage level from ts peers and at a dfferent rate of deceleraton resultng n complete loss of process control. Data processng and control equpment s also very senstve to voltage dps and can suffer from data loss and extended downtme. There are two man causes of voltage dps: startng of large loads ether on the affected ste or by a consumer on the same crcut and faults on other branches of the network. When heavy loads are started, such as large drves, the startng current can be many tmes the normal runnng current. Snce the supply and the cablng of the nstallaton are dmensoned for normal runnng current the hgh ntal current causes a voltage drop n both the supply network and the nstallaton. The magntude of the effect depends on how strong the network s, that s, how low the mpedance s at the pont of common couplng (PCC) and on the mpedance of the nstallaton cablng. Dps caused by startng currents are characterzed by beng less deep and much longer than those caused by network faults typcally from one to several seconds or tens of seconds, rather than less than one second. The extent of a voltage dp at one ste due to a fault n another part of the network depends on the topology of the network and the relatve source mpedances of the fault, load and generators at ther common pont of couplng. The duraton of the dp depends on the tme taken for the protectve crcuts to detect and solate the fault and s usually of the order of a few hundred of mllseconds. Snce faults - 8 -

15 2. POWER QUALITY IN ELECTRICAL SYSTEMS can be transtory, for example when caused by a tree branch fallng onto a lne, the fault can be cleared very soon after t has occurred. If the crcut were to be permanently dsconnected by the protecton equpment then all consumers on the crcut would experence a blackout untl the lne could be checked and reconnected. Autoreclosers can help to ease the stuaton, but also cause an ncrease n the number of dps. An autorecloser attempts to reconnect the crcut a short tme (less than 1 second) after the protecton equpment has operated. If the fault has cleared, the autoreclose wll succeed and power s restored. Loads on that crcut experence a 100 % dp between dsconnecton and autoreclose whle other loads see a smaller, shorter dp between the fault occurrng and beng solated, as dscussed above. If the fault has not cleared when the autorecloser reconnects, the protectve equpment wll operate agan; the process can be repeated accordng to the program set for the partcular autorecloser. Each tme t reconnects the faulty lne another dp results, so that other consumers can experence several dps n seres. Utlty performance n deregulated markets s partly - n some countres, such as UK, solely - judged on the average customer mnutes lost, takng nto account nterruptons exceedng, typcally, one mnute. Mnmzng ths statstc has resulted n the wdespread applcaton of autoreclosers and an ncrease n the probablty of dps. In other words, long term avalablty has been maxmzed but at the expense of qualty. Electronc equpment power supples, such as those used n personal computers (PC) and programmable logc controllers (PLC) employ a reservor capactor to smooth out the peaks of the full wave rectfed waveform, so they should be nherently reslent to short duraton dps. The larger the capactor, and the greater the dfference between the stored capactor voltage and the mnmum requred for the nternal voltage converters to operate, the better the reslence wll be. Desgners wll always try to reduce the sze of the capactor to a mnmum to reduce sze, weght and cost whle ensurng that the charge stored s just suffcent at mnmum voltage and maxmum load. For good dp reslence a much larger capactor s requred, at least twce as large to enable the equpment to rde through one cycle, and 100 tmes as large f a one-second rde through was requred. An alternatve desgn strategy s to keep the mnmum nput voltage as low as possble to maxmze the hold up tme of the system. Ths s the approach taken, by default, n equpment desgned to work over a wde range of voltage. For shallow dps, where there s consderable retaned voltage, there are several establshed automatc voltage regulator technologes ncludng electro-mechancal and electromagnetc devces. Because there s no need for stored energy, these devces can be used for long duraton events such as under and over voltage. Where heavy loads or deep dps are concerned a Dynamc Voltage Restorer s used. Ths devce s seres coupled to the load and - 9 -

16 2. POWER QUALITY IN ELECTRICAL SYSTEMS generates the mssng part of the supply; f the voltage dps to 70 %, the Restorer generates the mssng 30 %. Voltage Restorers are normally expected to support the load for a short perod and may use heavy-duty batteres, super capactors or other forms of energy storage such as hgh-speed flywheels, hence they cannot be used to correct long term under and over voltage Power Qualty Evaluaton Systematc procedures for evaluatng power qualty concerns can be developed but they must nclude all levels of the system, from the transmsson system to the end user facltes. Power qualty problems show up as mpacts wthn the end user faclty but may nvolve nteracton between all levels of the system. A consstent set of defntons for dfferent types of power qualty varatons s the startng pont for developng evaluaton procedures. The defntons permt standardzed measurements and evaluatons across dfferent systems. A data analyss system for power qualty measurements should be able to process data from a varety of nstruments and support a range of applcatons for processng data. Wth contnuous power qualty montorng, t s very mportant to be able to summarze varatons by means of tme trends and statstcs as well as characterze ndvdual events. Many nstruments and on-lne montorng equpment now nclude the capablty to sample waveforms and perform FFT calculatons. The capabltes of these nstruments vary wdely and the user must be careful that the accuracy and nformaton obtaned s adequate for the nvestgaton. The followng are some basc requrements for harmonc measurements used to nvestgate a problem:. Capablty to measure both voltage and current smultaneously so that harmonc power flow nformaton can be obtaned;. Capablty to measure both magntude and phase angle of each harmonc component;. Synchronzaton and a samplng rate matchng the correct and accurate measurement of both harmonc components and transent phenomena; v. Capablty to characterze the statstcal nature of harmonc dstorton levels (harmoncs levels change wth changng load and/or system condtons). Harmonc dstorton s a contnuous phenomena. It can be characterzed at a pont n tme by the frequency spectrums of the voltages and currents. However, for proper representaton, measurements over a perod of tme must be made and the statstcal characterstcs of the harmonc components and the total dstorton determned

17 2. POWER QUALITY IN ELECTRICAL SYSTEMS Research and Standardzaton actvty In the feld of voltage qualty, an ntense research actvty s conducted at nternatonal level; ths work gves rse to very mportant pre-standardzaton documents, that are often taken as bass for the development of the nternatonal standards. In the followng the most mportant workng groups and standardzaton commttees are reported, as well as ther actvty summarzed. CIGRE/CIRED Jont Workng Group on Voltage Qualty: Ths Jont Workng Group has carred out snce many years research actvty on the man aspects of voltage qualty,.e.: characterzaton of the varous types of low frequency electromagnetc dsturbances (harmoncs, flcker, voltage dps and swells), crtera for ther evaluaton, measurng methods; assessment of crtera to establsh adequate lmts for these dsturbances; mtgaton methods, cost analyss for the varous mtgaton methods and/or for ncreasng the mmunty of the sensble equpment. Recently some reorganzaton affected the actvty of the group but two mportant tems on the crtera for defnng voltage qualty held: characterzaton methods for assessng voltage qualty; qualty ndces and measurement protocols. The fnal WG Report presents power qualty data gathered from several dfferent countres across a number of montorng ponts over a number of years. The report provdes gudance on the key factors that need to be consdered when gatherng and presentng data. In so dong the report consders the benefts of consstency but recognzes the nherent dfferences between dfferent electrcal systems and dfferent power qualty objectves. The report develops the case for a consstent set of power qualty ndces and objectves that can be seen as the outer envelope of performance for each power qualty parameter. Relevant power qualty ndces are prerequstes for assessng ste and system performance wth respect to power qualty. Such ndces wll eventually facltate the task of system operators wth ther oblgaton to routnely report power qualty performance. Some ste ndces have already been defned n standards, but others are stll mssng - n partcular for hgh and extra-hgh voltage (HV-EHV) systems. Snce system operators are at rsk of beng exposed to penalty payments for excursons n qualty beyond the objectve values t s mportant that the objectves are seen not only as achevable but also as beng cost effectve for all customers. Ths adds to the ncentve for havng well defned and recognzed power qualty ndces. Optmzng the power qualty performance of the electrcal system s one of the roles of a system operator, the role of the regulator s to ensure that ths s carred out n a cost-effectve manner n that f customers expect power qualty to be an ntrnsc characterstc of the product they also want t at the lowest prce. Recognzng that hstorcally the electrcal systems n dfferent countres have been desgned n dfferent ways to cater for natonal/regonal varatons, such as dfferent commercal or clmatc

18 2. POWER QUALITY IN ELECTRICAL SYSTEMS condtons, t s essental that any sets of nternatonally agreed power qualty objectves also recognze these dfferences. IEEE Dstrbuton Subcommttee - Workng Group on Dstrbuton Voltage Qualty: Ths Workng Group carres out, manly n USA, a research and development actvty smlar to that of the CIGRE/CIRED Workng Group, wth whch t has a strct co-operaton. The IEEE Subcommttee has prepared some mportant pre-standards, two of them appear partcularly mportant:. IEEE 1159 [2], whch defnes ndces and crtera for the qualty level of the electrc energy;. IEEE 519 [4], regardng the crtera for checkng the harmoncs content. Both the standards refer to a thrd document, the IEEE 1459 [3]. It lsts the mathematcal expressons that were used n the past, as well as new expressons, and explans the features of the new defntons. The program of future work ncludes manly the revson and possble extenson of the above documents. EURELECTRIC/UNIPEDE and UIE Experts Groups: In the feld of voltage qualty mportant pre-standardzaton actvtes are also conducted wthn EURELECTRIC (Unon of the Electrcty Industry) / UNIPEDE (Internatonal Unon of Producers and Dstrbutors of Electrc Energy) and UIE (Internatonal Unon for the Applcaton of the Electrcty). Wth reference to UNIPEDE, the actvty has been carred out by the Expert Group Characterstcs of the product electrcty and electromagnetc compatblty of the Specfc Commttee on Standardzaton. Wth reference to UIE, the actvty s carred out by Workng Group Power Qualty, whch, at present, n cooperaton wth the above CIGRE/CIRED Workng Group, s preparng a Gude about the varous aspects of voltage qualty: types of dsturbances and relevant standards; voltage dps and short nterruptons; voltage dstorton; voltage unbalance; flcker; transent and temporary overvoltages. The real standardzaton actvty s carred out at the nternatonal level by IEC, at the European level by CENELEC, at the natonal level by CEI. IEC Subcommttee Electromagnetc compatblty - Low frequency phenomena : Ths IEC Subcommttee has prepared a seres of standards whch are of nterest for the defnton of voltage qualty. They can be classfed as follows: seres IEC x: standards for the defnton of the electromagnetc envronments and of the low frequency compatblty levels [7]; seres IEC x: standards for the lmtaton of the low frequency dsturbances produced by the equpment connected to the dstrbuton network [9];

19 2. POWER QUALITY IN ELECTRICAL SYSTEMS standards [13] and [14]: standards relatng to the nstrumentaton and to the measurng technques for the flcker and the harmoncs; standards [12]: standards relatng to the mmunty of equpment to low frequency conducted dsturbances. Wthn the Subcommttee, a specfc Workng Group on voltage qualty was set up to prepare a standard defnng detaled specfcatons for the nstrumentaton and the measurng methodologes. The work led to [15], whch specfes these measurng aspects for the varous parameters characterzng voltage qualty. CENELEC - TC 210 Electromagnetc compatblty : ts actvty n the feld of voltage qualty essentally conssts n the transposton nto European standards of the IEC standards descrbed n the prevous clause. CENELEC BTTF 68-6 Physcal characterstcs of electrcal energy : an ad-hoc Task Force of the Bureau Technque prepared, on the bass of a document avalable wthn UNIPEDE, the standard [1], publshed n ts frst edton n 1994: presently ths s the most mportant techncal reference n Europe for the regulaton of voltage qualty suppled n medum and low voltage publc dstrbuton networks. Ths standard has also been adopted n Italy by the Italan Electrotechncal Commttee as CEI The standard EN was not specfcally developed n relaton to the European Drectve 96/92/EC regardng the lberalzaton of the electrc energy market, but t was conceved as a voluntary techncal standard for the defnton of voltage qualty at the termnals of the energy supply to the medum and low voltage users, as a consequence of the European Drectve 85/374/EEC, whch consders the electrcal energy as a product. The problem of voltage qualty for the hgh voltage users and for the other ponts of energy exchange s outsde the scope of EN Consderng the partcular mportance of ths standard, ts content s reported n the followng secton Basc defntons of voltage parameters In standard [1] several voltage parameters are defned. In the followng the most mportant ones are reported: Supply voltage the RMS value of the voltage at a gven moment at the pont of common couplng, measured over a gven tme nterval. Nomnal voltage of the system (U n ) the voltage by whch a system s desgnated or dentfed and to whch certan operatng characterstcs are referred. Declared supply voltage (U c ) s normally the nomnal voltage U n of the system. If, by agreement between the suppler and the user, a voltage dfferent from the nomnal voltage s appled to the termnal, then ths voltage s the declared supply voltage U c

20 2. POWER QUALITY IN ELECTRICAL SYSTEMS Normal operatng condton the condton of meetng load demand, system swtchng and clearng faults by automatc system protecton n the absence of exceptonal condtons due to external nfluences or major events. Voltage varaton s an ncrease or decrease of voltage, due to varaton of the total load of the dstrbuton system or a part of t. Flcker mpresson of unsteadness of vsual sensaton nduced by a lght stmulus, the lumnance or spectral dstrbuton of whch fluctuates wth tme. Flcker severty ntensty of flcker annoyance defned by the UIE-IEC flcker measurng method and evaluated by the followng quanttes: Short term severty (P st ) measured over a perod of ten mnutes; Long term severty (P lt ) calculated from a sequence of 12 P st values over a two-hour nterval. Supply voltage dp a sudden reducton of the supply voltage to a value between 90% and 1% of the declared voltage U c, followed by a voltage recovery after a short perod of tme. Conventonally the duraton of a voltage dp s between 10 ms and 1 mn. The depth of a voltage dp s defned as the dfference between the mnmum RMS voltage durng the voltage dp and the declared voltage. Voltage changes whch do not reduce the supply voltage to less than 90% of the declared voltage U c are not consdered to be dps. Supply nterrupton s a condton n whch the voltage at the supply termnals s lower than 1% of the declared voltage U c. A supply nterrupton s classfed as: prearranged n order to allow the executon of scheduled works on the dstrbuton system, when consumers are nformed n advance, or accdental, caused by permanent (a long nterrupton) or transent (a short nterrupton) faults, mostly related to external events, equpment falures or nterference. Temporary power-frequency overvoltages have relatvely long duraton, usually of a few power frequency perods, and orgnate manly from swtchng operatons or faults, e.g. sudden load reducton, or dsconnecton of short crcuts. Transent overvoltages are oscllatory or non-oscllatory, hghly damped, short overvoltages wth a duraton of a few mllseconds or less, orgnatng from lghtnng or some swtchng operatons, for example at swtch-off of an nductve current. Harmonc voltage a snusodal voltage wth a frequency equal to an nteger multple of the fundamental frequency of the supply voltage. Harmonc voltages can be evaluated: ndvdually by ther relatve ampltude U h related to the fundamental voltage U 1, where h s the order of the harmonc Voltage Characterstcs of Publc Dstrbuton Systems; globally, usually by the total harmonc dstorton factor THD. Interharmonc voltage s a snusodal voltage wth frequency between the harmoncs,.e. the frequency s not an nteger multple of the fundamental. Voltage unbalance s a condton where the RMS value of the phase voltages or the phase angles between consecutve phases n a three-phase system are not equal. The standard [1] consders two groups of parameters characterzng voltage qualty; for the frst group lmt values are ndcated, whereas for the second only ndcatve values. Frequency, ampltude of the voltage (slow varatons), rapd varatons of the voltage, flcker (voltage fluctuatons), harmonc dstortons, nterharmoncs, three phase voltage unbalance and level of

21 2. POWER QUALITY IN ELECTRICAL SYSTEMS communcaton sgnals njected on the network, are all parameters belongng to the frst group. The second group ncludes the followng other parameters: voltage dps and swells; short and long nterruptons; transent and temporary overvoltages. The standard does not contan detaled nformaton on the nstrumentaton and on the measurng technques to be adopted to assess the conformty of voltage qualty. However, t gves general suggestons for the varous parameters to be measured on the crtera for choosng the value (average value, RMS value, peak value, etc) that characterzes the parameter to be measured. Even the statstcal method of evaluaton s suggested, as far as the ndcaton of the confdence level that a certan value s not exceeded (e.g.: 95%, 99%, 100%), the tme nterval necessary to obtan a sngle measurement (10 ms, 3 s, 10 s, 10 mn.) and the observaton perod (one day, one week, one year). It does not apply under abnormal operatng condtons, such as: condtons arsng as a result of a fault; n case of falure of a customer s nstallaton or equpment to comply wth the relevant standards or wth the techncal requrements for the connecton of loads; n the event of the falure of a generator nstallaton to comply wth relevant standards or wth the techncal requrements for nterconnecton wth an electrcty dstrbuton system; n exceptonal stuatons outsde the electrcty suppler s control, n partcular: exceptonal weather condtons and other natural dsasters, thrd party nterference, actons of publc authortes, ndustral acton (subject to legal requrements), power shortages resultng from external events. Actually requrements are not partcularly rgorous for the suppler. In fact the numerous stuatons n whch the standard does not apply can excuse the majorty of outages and voltage dsturbance events that occur n practce. Thus, many supplers nterpret the requrements of EN as prncpally nformatve and clam no responsblty when the lmts are exceeded. On the other hand, the consumer s pont of vew s usually totally dfferent regardng the lmts gven as requrements that must be guaranteed by the suppler. However, as mentoned before, for many consumers, even fulfllng the requrements of [1] does not assure a satsfactory level of PQ. In such cases the level of PQ requred must be defned n a separate agreement between suppler and consumer Overvew of power qualty ndces Harmonc Components Obtanng harmonc ndces conssts of provdng the spectrum of voltage or current over a gven wndow of tme; a ste ndex from the spectra over a gven perod; and eventually a system ndex from the sngle ste ndces. Varous methods for obtanng the spectrum

22 2. POWER QUALITY IN ELECTRICAL SYSTEMS are dscussed n the techncal lterature, but the method almost exclusvely used n power qualty montorng s the Fourer transform. A number of nternatonal standard documents defne the measurement process, ncludng [13] and [15]. The method proceeds as follows: obtan the spectrum over a 10-cycle (50 Hz systems) or 12-cycle (60 Hz systems) wndow. The wndow shall be synchronzed to the actual frequency durng the measurement; the spectra (RMS) are combned to a spectrum over a 3-second nterval (150 cycles for 50 Hz systems and 180 cycles for 60 Hz systems) and the so obtaned values are referred to as very short tme ndces (U h,vs ); the 3-second values are combned to a 10-mnute value and referred to as short tme ndces (U h,sh ); 3-second and 10-mnute values are evaluated over a one-day or a one-week perod dependng on the ndex. The 95%, 99% or maxmum values of the dstrbutons are used as ste-ndces. Other publcatons propose more specfc ndces such as: Techncal report [11]: The greatest 95 % probablty daly value of U h,vs (RMS value of ndvdual harmonc components over "very short" 3 s perods); The maxmum weekly value of U h,sh (RMS value of ndvdual harmoncs over "short" 10 mn perods); The maxmum weekly value of U h,vs. For measurements t refers to [13]. The mnmum measurement perod should be one week. Standard [15] also refers to [13] for measurements, more specfcally to class 1, 10/12- cycle gapless harmonc sub-group measurement. The standard does not specfy ndces, but varous ndces are gven as gudelnes for contractual applcatons The number, or percentage, of values durng the nterval that exceed contractual values mght be counted; The worst-case values mght be compared to contractual values (the measurement nterval mght be dfferent for ths possblty, for example one year); One or more 95 % (or other percentage) probablty weekly values for 10-mnute values, 95 % (or other percentage) probablty daly values for 3-sec tme nterval values, expressed n percent, mght be compared to contractual values. A mnmum assessment perod of one week s recommended for 10-mn values, and daly assessment of 3-sec values for at least one week. Standard [1], stpulates that durng each perod of one week, the percentle 95% of the 10- mn mean RMS value (U h,sh ) of each ndvdual harmonc voltage s the qualty ndex to be compared to the relevant

23 2. POWER QUALITY IN ELECTRICAL SYSTEMS voltage characterstc. Other regonal or natonal standards and gudelnes also recommend ndces that are often smlar to those mentoned above. Flcker The flckermeter algorthm as defned n [14] results n: a 10-mnute short-term flcker severty - P st. Ths value s obtaned from a statstcal analyss of the nstantaneous flcker value n a way whch models ncandescent lamps and our observaton of lght ntensty varatons. From the 10-mnute value, a 2-hour long-term flcker severty - P lt s calculated. Indces of flcker severty (P st and P lt ) are expressed n per unt of the rrtablty threshold of flcker, that s the level of flcker consdered rrtable by a sgnfcant porton of the people nvolved n the tests. Evaluaton technques mght be agreed between partes: the number or percentage of values durng the nterval that exceed contractual values mght be counted, as well as 99 % probablty weekly values for P st, or 95 % probablty weekly value for P lt, mght be compared to contractual values. Unbalance Only the fundamental components shall be used: all harmonc components should be elmnated by usng DFT algorthm. The processng s defned smlar as the above harmonc ndces: from 10-cycle (50 Hz) and 12-cycle (60 Hz), to 3-second ntervals, to 10-mnute ntervals. For unbalance also 2-hour values (obtaned by combnng 10-mnute values) are used. The whole measurement and evaluaton procedure s defned n detal n [15]. Ths standard suggests that 10-mn and/or 2-hr values be assessed as follows:. The number of values durng the measurement nterval that exceed contractual values mght be counted;. the worst-case values mght be compared to contractual values (the measurement nterval mght be dfferent for ths possblty, for example one year);. one or more 95 % (or other percentage) probablty weekly values, NPS expressed as a percentage of PPS, mght be compared to contractual values. In standard EN the unbalance ndex s the 95 % of the 10-mn mean RMS values of the negatve phase sequence component of the supply voltage to be assessed durng each perod of one week. The voltage lmts set n ANSI Standard C84.1 at the pont of use are at ± 10%, deratng motor capacty at levels of unbalance greater than 1% and not exceedng 5%. The deratng s based on the thermal effects on motors, and are therefore presumed to be related to long-tme measurements rather than short-tme measurements. The measurement specfed s the dfference between the average of the three phase magntudes and the voltage that dffers the most from that average, dvded by the average

24 2. POWER QUALITY IN ELECTRICAL SYSTEMS Voltage Dps The frst nternatonal defnton and measurement method for the most common characterzaton of voltage dps n terms of magntude and duraton s provded n [15]. For the measurement of dps, such standard states that the basc measurement of a voltage dp and swell shall be the value of the RMS voltage measured over one cycle and refreshed each half cycle. From the RMS voltage as a functon of tme two basc characterstcs can be determned: retaned voltage or the dp depth; duraton. A voltage dp s characterzed by a par of data, ether retaned voltage and duraton or depth and duraton: the retaned voltage s the lowest value measured on any channel durng the dp; the depth s the dfference between the reference voltage and the retaned voltage expressed n % of the reference voltage; the duraton of a voltage dp s the tme dfference between the begnnng and the end of the voltage dp. The choce of a dp threshold s essental for determnng the duraton of the event. Ths choce of threshold s also mportant for countng events, as events are only counted as voltage dps when the RMS voltage drops below the threshold. Dp threshold can be a percentage of ether nomnal or declared voltage, or a percentage of the sldng voltage reference, whch takes nto account the actual voltage level pror to the occurrence of a dp. The user shall declare the reference voltage n use. Voltage dp envelopes may not be rectangular hence, for a gven voltage dp, the measured duraton depends on the selected dp-threshold value. The shape of the envelope may be assessed usng several dp thresholds set wthn the range of voltage dp and voltage nterrupton threshold detecton. A number of other characterstcs for voltage dps are mentoned n an annex to [15] ncludng phase angle shft, pont-on-wave, three-phase unbalance, mssng voltage and dstorton durng the dp. The use of addtonal characterstcs and ndces may gve addtonal nformaton on the orgn of the event, on the system and on the effect of the dp on equpment. Even though several of these terms are used n the power-qualty lterature there s no consstent set of defntons. Document [8] also refers to [15] for measurement, but ntroduces a number of addtonal recommendatons for calculatng voltage-dp ndces. Recommended values are 90% and 91% for dp-start threshold and dp-end threshold, respectvely, and 10% for the nterrupton threshold. Dps nvolvng more than one phase should be desgnated as a sngle event f they overlap n tme. The most commonly-referred to ndex s the System Average RMS varaton Frequency Index or SARFI

25 2. POWER QUALITY IN ELECTRICAL SYSTEMS The term RMS varaton s used n US lterature to ndcate all events n whch the RMS voltage devates sgnfcantly (typcally seen as more than 10%) from ts nomnal value. Ths ncludes voltage dps, voltage swells and long nterruptons. The SARFI X ndex (where X s a number between 0 and 100%) gves the number of events per year wth a duraton between 0,5 cycle and 1 mnute and a retaned voltage less than X%. Thus SARFI 70 gves the number of events wth retaned voltage less than 70%. Strctly speakng, SARFI values are obtaned as a weghted average over all montor locatons wthn a supply network or wthn part of the supply network. However the term s also used to refer to the event frequency at one locaton. By usng the weghtng factors, more weght can be gven to locaton wth more or more mportant load. The weghtng factors are n most cases taken to be equal for all locatons. Indces used for transmsson nterrupton reportng dffer sgnfcantly from utlty to utlty. The ndces used can however be dvded nto the followng categores: Number of events: actual number of events and the average number of events over the reportng perod,.e. the frequency of events; Duraton of events: average total duraton of events over the reportng perod and average tme to restore supply per nterrupton at each supply pont. The avalablty of the supply s the converse of the duraton and t gves an ndcaton of the relatve rsk of nterruptons; Severty of events: severty of the nterrupton events over the reportng perod (.e. the sze of load affected) and ndces estmatng the cost mpact per event Standard Measurement methods of PQ parameters Requrements of EN The correct operaton of electrcal equpment requres a supply voltage that s as close as possble to the rated voltage. Even relatvely small devatons from the rated value can cause operaton at reduced effcency, or hgher power consumpton wth addtonal losses and shorter servce lfe of the equpment. Sometmes prolonged devatons can cause operaton of protecton devces, resultng n outages. Of course, the correct operaton of equpment also depends on many other factors, such as envronmental condtons and proper selecton and nstallaton. Investgaton of the ndependent nfluence of each supply voltage parameter on equpment operaton s easly performed, but when parameters vary smultaneously the stuaton s much more complex. In some cases, after detaled analyss of the effects of each of the dfferent voltage parameters, results can be supermposed n order to estmate the total nfluence of many parameters. The ncreased concern for power qualty has resulted n sgnfcant advances n montorng equpment that can be used to characterze dsturbances and power qualty

26 2. POWER QUALITY IN ELECTRICAL SYSTEMS varatons. In partcular, measurement and testng of supply voltage qualty accordng to EN requres specalzed apparatus and measurng methods montorng, contnuously over 7 days, the followng parameters: voltage n three phases frequency total harmonc dstorton factor THD U voltage unbalance factor, whch s a multple of postve and negatve sequence voltage components fast and slow voltage varatons, whch are defned as short term (P st ) and long term (P lt ) flcker severty factors. Ths arrangement also enables measurement of voltage dps and outages, ts frequency and duraton. The measured parameters are processed and recorded as 10 mnute tme-segments (1008 segments over 7 days). For each segment the mean value of the measured parameter s calculated. After the 7-day recordng perod a so-called ordered dagram s produced, whch shows the sum of the duraton of a gven dstorton level n the observed tme perod. (for frequency measurement, the duraton of each sngle segment s 10 seconds). Requrements of IEC seres Methods for measurement and nterpretaton of results for power qualty parameters n 50/60 Hz a.c. power supply systems are defned n [15]. Measurement methods are descrbed for each relevant type of parameter n terms that wll make t possble to obtan relable, repeatable and comparable results regardless of the complant nstrument beng used and regardless of ts envronmental condtons. Ths standard addresses methods for measurements carred out at the montored pont of the system. Measurement of parameters covered by ths standard s lmted to those phenomena that can be conducted n a power system. These nclude the voltage and/or current parameters, as approprate. The power qualty parameters consdered n ths standard are power frequency, magntude of the supply voltage, flcker, supply voltage dps and swells, voltage nterruptons, transent voltages, supply voltage unbalance, voltage and current harmoncs and nterharmoncs, mans sgnallng on the supply voltage and rapd voltage changes. Dependng on the purpose of the measurement, all or a subset of the phenomena on ths lst may be measured. The effects of transducers beng nserted between the power system and the nstrument are acknowledged but not addressed n detal n ths standard. Precautons on nstallng montors on lve crcuts are addressed. Measurements can be performed on sngle-phase or polyphase supply systems. Dependng on the context, t may be necessary to measure voltages between phase

27 2. POWER QUALITY IN ELECTRICAL SYSTEMS conductors and neutral (lne-to-neutral) or between phase conductors (lne-to-lne) or between neutral and earth. The basc measurement tme nterval for parameter magntudes (supply voltage, harmoncs, nterharmoncs and unbalance) shall be a 10- cycle tme nterval for 50 Hz power system or 12-cycle tme nterval for 60 Hz power system. Measurement tme ntervals are aggregated over 3 dfferent tme ntervals. The aggregaton tme ntervals are: 3-s nterval (150 cycles for 50 Hz nomnal or 180 cycles for 60 Hz nomnal), 10-mn nterval, 2-h nterval. Aggregatons are performed by usng the square root of the arthmetc mean of the squared nput values. Three categores of aggregaton are necessary:. Cycle aggregaton - The data for the 150/180-cycle tme nterval shall be aggregated from ffteen 10/12-cycle tme ntervals. Ths tme nterval s not a "tme clock" nterval; t s based on the frequency characterstc.. From cycle to tme-clock aggregaton: The 10-mn value shall be tagged wth the absolute tme. The tme tag s the tme at the end of the 10-mn aggregaton. If the last 10/12-cycle value n a 10-mn aggregaton perod overlaps n tme wth the absolute 10-mn clock boundary, that 10/12-cycle value s ncluded n the aggregaton for ths 10-mn nterval. On commencement of the measurement, the 10/12-cycle measurement shall be started at the boundary of the absolute 10-mn clock, and shall be re-synchronzed at every subsequent 10-mn boundary.. Tme-clock aggregaton: The data for the 2-h nterval shall be aggregated from twelve 10-mn ntervals. Durng a dp, swell, or nterrupton, the measurement algorthm for other parameters (for example, frequency measurement) mght produce an unrelable value. The flaggng concept therefore avods countng a sngle event more than once n dfferent parameters (for example, countng a sngle dp as both a dp and a frequency varaton) and ndcates that an aggregated value mght be unrelable. Flaggng s only trggered by dps, swells, and nterruptons. The detecton of dps and swells s dependent on the threshold selected by the user, and ths selecton wll nfluence whch data are "flagged"

28 2. POWER QUALITY IN ELECTRICAL SYSTEMS Table 2-1. Standard measurement methods of the voltage qualty parameters. Power frequency The frequency readng shall be obtaned every 10-s. As power frequency may not be exactly 50 Hz or 60 Hz wthn the 10-s tme clock nterval, the number of cycles may not be an nteger number. The fundamental frequency output s the rato of the number of

29 2. POWER QUALITY IN ELECTRICAL SYSTEMS ntegral cycles counted durng the 10-s tme clock nterval, dvded by the cumulatve duraton of the nteger cycles. Before each assessment, harmoncs and nterharmoncs shall be attenuated to mnmze the effects of multple zero crossngs. The measurement tme ntervals shall be non-overlappng. Indvdual cycles that overlap the 10-s tme clock are dscarded. Each 10-s nterval shall begn on an absolute 10-s tme clock, ±20 ms for 50 Hz or ±16,7 ms for 60 Hz. Magntude of the supply voltage The measurement shall be the RMS value of the voltage magntude over a 10-cycle tme nterval for 50 Hz power system or 12-cycle tme nterval for 60 Hz power system. Every 10/12-cycle nterval shall be contguous wth, and not overlap, adjacent 10/12-cycle ntervals. Classes of measurement performance For each parameter measured, two classes of measurement performance are defned. Class A performance Ths class of performance s used where accurate measurements are necessary, for example, for contractual applcatons, verfyng complance wth standards, resolvng dsputes, etc. Any measurements of a parameter carred out wth two dfferent nstruments complyng wth the requrements of class A, when measurng the same sgnals, wll produce matchng results wthn the specfed uncertanty. To ensure that matchng results are produced, class A performance nstrument requres a bandwdth characterstc and a samplng rate suffcent for the specfed uncertanty of each parameter. Class B performance Ths class of performance may be used for statstcal surveys, trouble-shootng applcatons, and other applcatons where low uncertanty s not requred. For each performance class the range of nfluencng factors that shall be compled wth s specfed n [15]. Users shall select the class of measurement performance takng account of the stuaton of each applcaton case. A measurement nstrument may have dfferent performance classes for dfferent parameters. The nstrument manufacturer should declare nfluence quanttes whch are not expressly gven and whch may degrade performance of the nstrument. Voltage Harmoncs Table 2-2: provdes a summary comparson of harmonc ndces between varous standards and gudelnes. It shows that n most cases the reference standard to perform harmonc measurements s [13]. Ths part of IEC s applcable to nstrumentaton ntended for measurng spectral components n the frequency range up to 9 khz whch are supermposed on the fundamental of the power supply systems at 50 Hz and 60 Hz

30 2. POWER QUALITY IN ELECTRICAL SYSTEMS For practcal consderatons, ths standard dstngushes between harmoncs, nterharmoncs and other components above the harmonc frequency range, up to 9 khz. Ths standard defnes the measurement nstrumentaton ntended for testng ndvdual tems of equpment n accordance wth emsson lmts gven n certan standards (for example, harmonc current lmts as gven n [10]) as well as for the measurement of harmonc currents and voltages n actual supply systems. Instrumentaton for measurements above the harmonc frequency range, up to 9 khz s tentatvely defned Practcally, the most common ndex for harmonc voltage s the so-called short tme or 10- mn value (U h,sh ). It s used manly for voltage characterstcs and the level of harmoncs to be compared wth the objectves s usually the value correspondng to 95% probablty of weekly statstcs. Instruments for the harmonc and nterharmonc emsson measurement or for measurements above the harmonc frequency range up to 9 khz are consdered n the IEC standard. Strctly speakng, harmonc measurements can be performed only on a statonary sgnal; fluctuatng sgnals cannot be descrbed correctly by ther harmoncs only. However, n order to obtan results that are nter-comparable, a smplfed and reproducble approach s gven for fluctuatng sgnals. Two classes of accuracy (I and II) are consdered, to permt the use of smple and low-cost nstruments, consstent wth the requrements of the applcaton. For emsson tests, the upper class I s requred f the emssons are near to the lmt values. New desgns of nstrument are lkely to use the dscrete Fourer transform (DFT), normally usng a fast algorthm called fast Fourer transform (FFT). Therefore the standard consders only ths archtecture but does not exclude other analyss prncples. The man nstrument for harmonc frequency measurements comprses: nput crcuts wth ant-alasng flter; A/D-converter (ncludng sample-and-hold unt); synchronzaton and wndow-shapng unt; DFT-processor provdng the Fourer coeffcents a m and b m. The nstrument s complemented by the specal parts devoted to current assessment and/or voltage assessment. For full complance wth ths standard, the wndow wdth shall be 10 (50 Hz systems) or 12 (60 Hz systems) perods wth rectangular weghtng. Hannng weghtng s allowed only n the case of loss of synchronzaton. Ths loss of synchronzaton shall be ndcated on the nstrument dsplay and the data so acqured shall be flagged. The tme wndow shall be synchronzed wth each group of 10 or 12 cycles accordng to the power system frequency of 50 Hz or 60 Hz. The tme between the leadng edge of the frst samplng pulse and the leadng edge of the (M+1) th samplng

31 2. POWER QUALITY IN ELECTRICAL SYSTEMS Table 2-2. Summary comparson of harmoncs ndces between dfferent standards and reference documents. Harmonc voltage ndces Internatonal standard or gudelnes Regonal or natonal standards and gudelnes Standard /document IEC IEC EN ANSI/IEEE 519 NRS EDF Emeraude Contract ER G5/4 H. Q. Voltage Characterstcs Status Techncal report type Internatonal Std. European Std. ANSI std. recommended practce South Afrcan Std. (France) PQ contract UK Natonal Std. Voluntary (Quebec) Purpose Indcatve plannng levels for emsson lmts Power qualty measurement methods Supply voltage characterstcs for publc networks Emsson lmts and system desgn methods Mnmum Std. used by the regulator Supply voltage characterstc Plannng levels for controllng emssons Supply voltage characterstc Very short tme ndces U h,vs 95% daly U h,vs X% as agreed Short tme ndces U h,s max. weekly U h,s X% as agreed U h,s + THD 95% weekly U h,s + THD 95% weekly U h,s + THD max U h,s + THD 95% weekly Other ndces U h,vs max. weekly 95% (no defnte ndces) U h1 mn + THD 95% weekly Perod for statstcal assessment One week mnmum At least one week or more One week Undefned One week mn. At least one week or more One week One week Measurement method IEC IEC IEC No specfc method Specfed method IEC Specfed method IEC pulse (where M s the number of samples) shall be equal to the duraton of the specfed number of cycles of the power system, wth a maxmum permssble error of ±0,03%. Instruments ncludng a phase-locked loop or other synchronzaton means shall meet the requrements for accuracy and synchronzaton for measurng at any sgnal frequency wthn a range of at least ±5% of the nomnal system frequency. However, for nstruments havng ntegrated supply sources, so that the source and measurement systems are nherently synchronzed, the requrement for a workng nput frequency range does not apply, provded the requrements for synchronzaton and frequency accuracy are met. The output shall provde the ndvdual coeffcents a m and b m of the DFT, for the current or voltage,.e. the value of each frequency component calculated. A further output, not necessarly from the DFT, shall provde the actve power P evaluated over the same tme wndow used for the harmoncs. For the harmonc emsson measurements accordng to [10], ths power shall not nclude the d.c. component. Flcker The mnmum measurement perod should be one week (see [14]). For flcker, ndces should be: 1. P st 99% weekly; 2. P lt 99% weekly. Standard [15] also refers to standard [14] for flcker measurement. Voltage dps, swells, and nterruptons shall cause P st and P lt

32 2. POWER QUALITY IN ELECTRICAL SYSTEMS Table 2-3. summary comparson of flcker ndces between dfferent standards and reference documents Flcker ndces Internatonal standard or gudelnes Regonal/ natonal standards and gudelnes Standard /document IEC IEC EN NRS EDF Emeraude Contract A2 ER P28 H. Q. Voltage Characterstcs Status Techncal report type Internatonal Std. European Std. South Afrcan Std. (France) PQ contract UK Natonal Std. Voluntary (Quebec) Purpose Indcatve plannng levels for emsson lmts Power qualty measurement methods Supply voltage characterstcs for publc networks Mnmum Std. used by the regulator Supply voltage characterstc Plannng levels for controllng emssons Supply voltage characterstc Short tme ndces P st 99% weekly P st 99% weekly or X% as agreed P st no further specfcaton Long tme ndces P lt 99% weekly P lt 95% weekly or X% as agreed P lt 95% weekly P st 95% weekly P lt no further specfcaton P lt no further specfcaton P lt 95% weekly Perod for statstcal assessment One week mnmum At least one week or more as agreed One week One week mn. At least one week or more Suffcent to capture full operatng cycle of load One week Measurement method IEC IEC IEC IEC IEC IEC 868 IEC output values to be flagged so that they can later be removed from statstcs. P st or P lt mght be consdered. The mnmum assessment perod should be one week. The mnmum measurement perod should be one week (see [14]). For flcker, ndces should be: 1. P st 99% weekly; 2. P lt 99% weekly. Standard [15] also refers to standard [14] for flcker measurement. Voltage dps, swells, and nterruptons shall cause P st and P lt Table 2-3 provdes a summary comparson of flcker ndces between varous standards and gudelnes. The most common reference for flcker measurement s bascally standard [14]. The 95% or 99% weekly values of P st or P lt ndces are mostly n use. Consderng that P lt and P st values are often correlated by a defnte or quas-constant factor related to the characterstcs of the dsturbng process, t may be questoned whether t s redundant specfyng both ndces. Unbalance Table 2-4 summarses the ndces relevant to negatve sequence voltage unbalance factor (U neg ). 10-mn values are most commonly n use. Although dfferent equatons may be used for calculatng voltage unbalance factor, results should be smlar for a gven ntegraton tme provded they consder negatve sequence voltage. Voltage dps Varous methods for reportng dps or sags have been proposed n lterature. They can be classfed n two categores: methods to characterze ste or system performance as such, and methods most sutable to estmate the compatblty between equpment and supply. Magntude-duraton table: Ste performance as well as system performance are often descrbed n the form of a voltage-dp table. Dfferent table formats are dscussed n [5]

33 2. POWER QUALITY IN ELECTRICAL SYSTEMS but only the so-called densty table s commonly used. The columns of the table represent ranges of voltage-dp duraton; the rows represent ranges of retaned voltage. The choce Table 2-4. summary comparson of flcker ndces between dfferent standards and reference documents. Voltage unbalance ndces Standard /document Status Purpose Very short tme ndces Short tme ndces Long tme ndces Perod for statstcal assessment Internatonal standard or gudelnes Regonal or natonal standards and gudelnes IEC Cgré 1992 paper EN NRS EDF Emeraude Contract A2 ER P29 Internatonal Std. Cgré work European Std. South Afrcan Std. (France) PQ contract UK Natonal Std. Power qualty measurement methods Assessng voltage qualty n relaton to harmoncs, flcker and unbalance Supply voltage characterstcs for publc networks Mnmum Std. used by the regulator Supply voltage characterstc Plannng levels for controllng emssons U neg,vs 95% daly U neg,s 95% weekly or as agreed U neg,s max. weekly U neg,s 95% weekly U neg,s 95% daly U neg,s no further specfcaton Max. of negatve sequence measured over 1 mnute U neg,l 95% weekly or as agreed One week mnmum A few days ncludng a week end One week One week mn. At least one week or more Suffcent to represent effect on rotatng plant H. Q. Voltage Characterstcs Voluntary (Quebec) Supply voltage characterstc U neg,l 95% weekly One week of the magntude and duraton ranges for voltage-dp tables s a pont of dscusson. Dfferent publcatons use dfferent values. Voltage-sag coordnaton chart: A method for reportng ste nformaton from event magntude and duraton s descrbed n [5] and [6]. The method results n the so-called voltage sag coordnaton chart. An example of such a chart s shown n Fgure 2-3. Ths fgure s the result of montorng from 6 years at 20 HV-stes. The chart, as defned n these standards, contans the performance of the supply at a gven ste, and the voltage tolerance of one or more devces. For the purpose of ths document only the supply performance part of the chart s of relevance. The chart gves the number of events per year (sags and nterruptons) as a functon of the severty of the event. For the example shown here there s on average 1 event per year droppng the voltage below 50% for 100 ms or longer. There s also on average 1 event per year more severe than 80%, 80 ms and on average 0.1 event per year below 70% for longer than 500 ms

34 2. POWER QUALITY IN ELECTRICAL SYSTEMS Fgure 2-3. Voltage-sag coordnaton chart Non conventonal parameters for PQ measurement Recently, the worldwde research actvty n the feld of both electrcal systems and electrc and electronc measurements has been focused on the detecton and localzaton of the sources of dsturbance wthn a network. The soluton to ths ssue s gettng more and more mportant n the management of electrc systems because the locaton of a dsturbng devce can be strctly related to the economcal and contractual aspects between utlty and customers. Relable results of power qualty montorng systems depend on the correctness of the theoretcal model of the electrc network and of the nstalled devces. Frst of all, the snusodal balanced steady-state condton can be assumed no longer for modellng the whole electrcal system, even though the 50 Hz steady state s consdered as usual workng condton for desgnng the devces and for the most popular smulaton and measurement technques. Consderng for example the electrc lnes model: the smple 4- pole tap parameters model used correctly for a 50 Hz short electrc lne (220 kv, 100 km), has to be swtched to a more complcated frequency-dependent model f the sgnal has a 2500 Hz frequency. The equvalent lne length n ths case s 5000 km. Hence, the model of the system has to be adapted n functon of the type of dsturbance under test. When the presence and the man characterstcs of a dsturbance affectng the network are known, ts propagaton along the lnes and loads has to be analyzed. In dong ths, the man ssue s the mathematcal representaton of the nteracton between the devce generatng the dsturbance and the system affected by ts effects. Unfortunately, n most cases the drect method s not correct: t would consst n analyzng the dsturbance contrbuton of the load under test by tself and then njectng the dsturbance nto the system to check ts behavour wthout the consdered load be nstalled. The nteractons

35 2. POWER QUALITY IN ELECTRICAL SYSTEMS have no sgnfcant effect on the measurement nstrumentaton, but can affect the measurement procedure and/or the results nterpretaton. The scentfc lterature proposes technques based on dfferent operatng prncples. No one of them, anyway, has been approved yet by the entre scentfc communty and by the governments, n facts each proposed approach can regster and show some crtcal aspects of the network under test but can be msleadng about other aspects. Some surveys or evaluatons about the locaton methods proposed n case of perodc dsturbances are reported n [17-24]. Instead of descrbng the theory standng besde each approach, t may be worthy to underlne ther common characterstc: most of the methods locatng a source of dsturbance n a network depend strongly on the measurement system chosen to mplement t. For example, let s consder the technques based on the evaluaton of the sgn of the harmoncs actve power contrbuton; the source of harmonc dsturbances s assumed to be the load f the harmonc actve power s negatve, otherwse t s assumed to be the electrc system f the harmonc actve power s postve. The approach s suggested by the standard [13] amng at recognzng the load responsble for the dstorton affectng the lne voltages, hence t s mplemented n numerous commercal nstruments desgned for Power Qualty montorng, even f the scentfc communty showed the technque to be weak, manly due to the fact that a load can be classfed as pollutng or polluted dependng on the operaton of all the other loads connected to the same power network. Anyway, the method based on the sgn of harmonc power components, compensated n ts wrong aspects, s the most popular for the locaton of perodc dsturbance sources. Hence, attenton must be focused on the metrologcal characterstcs of nstrumentaton used for the applcaton of the method. In partcular, the condtonng block,.e. voltage and current transducers, must be carefully chosen n order to avod that ther phase error, by varyng wth frequency, leads to wrong evaluaton of the harmonc components flow. On the bass of the sgn of the harmonc actve powers dsplayed by the nstrument the operator would then mstake n assessng responsbltes for the voltage power qualty degradaton. Often research groups present methods for the locaton of he source of dsturbances based on dstrbuted measurement systems; the analyss of data smultaneously captured n numerous ponts of the montored network gves much more nformaton on the system condtons. By the metrologcal pont of vew, dstrbuted measurements ntroduce new uncertanty sources respect to spot measurements, such as lack of synchronzaton and data transmsson delay

36 2. POWER QUALITY IN ELECTRICAL SYSTEMS 2.4. European Scenaro: standards, gudes and PQ level There s a worldwde trend of countres reformng ther power sectors: lberalzaton and prvatzaton have been ntroduced and a new approach s taken to the regulaton of the remanng network monopoles. Generally, the man objectves of power sector reform have been to mprove effcency and qualty levels. Regulators are assgned the task to attan objectves that are benefcal for socety, and these typcally nclude the promoton of hgh economc effcency and adequate levels of qualty. Relablty s the most mportant qualty feature n electrcty dstrbuton, n facts t s consdered the core value of electrcty servce provson. Any servce nterrupton temporarly ceases the provson of electrcty therewth drectly affects consumers. Network relablty means the contnuous avalablty of electrcty for the consumer. It s characterzed by the number of outages for a customer and the duraton of these outages. To most customers, t represents the most vsble and sensble ssue concernng the qualty of supply. Therefore many regulators prortze network relablty when startng regulatng qualty of supply. The stuaton of dstrbuton and transmsson networks s very dfferent, as the frst ones are characterzed wth many outages wth relatvely long duraton and affectng a lmted number of customers, whereas the latter ones are affected by rare outages, usually short n duraton but nvolvng many customers. The most nterruptons are caused n the low voltage (LV) grd, followed by medum voltage (MV) grds. A survey by the Italan regulator n 1998 showed that wthn Italy the MV grds are responsble for 85% of the total mnute lost, followed by the LV grd (12%) and the HV grds (3%). As mentoned above, whle EN gves general lmts for publc supply networks, varous European countres have addtonal rules governng supply condtons. Many of these natonal regulatons cover areas not ncluded n EN 50160, such as the maxmum permssble harmonc load to be connected to the PCC. The German natonal standard VDE 0100 states that the voltage parameters defned n DIN EN reflect extreme stuatons n the network and are not representatve of typcal condtons. In plannng networks the recommendatons of VDE 0100 should be followed: t gves maxmum values (per unt) for phase-angle controlled resstve loads (1700 VA sngle-phase, 3300 VA two-phase and 5000 VA balanced three-phase) and for uncontrolled rectfer loads wth capactve smoothng (300 VA sngle phase, 600 VA twophase and 1000 VA balanced three-phase). The equpment standard VDE 0838 (EN 60555) s also quoted. In Poland, the rules of electrcal energy dstrbuton establshed by the government gve the fundamental parameters of the supply voltage and do not refer to EN

37 2. POWER QUALITY IN ELECTRICAL SYSTEMS Addtonally, consumers are dvded nto sx groups, for whch separate, permssble total annual outage tmes are defned. The document also deals n detal wth varous economc aspects of the energy market, prncples of settlement between network and dstrbuton companes etc. In Italy there s an mportant document dealng wth the contnuty of suppled energy [16]. The Italan Regulatory Authorty for Electrcty and Gas (AEEG) has n fact set out a unform system of servce contnuty ndcators and has put n place a system of ncentves and penaltes n order to progressvely brng contnuty levels up to meet European standards. The Authorty has dvded the natonal terrtory nto 230 geographcal zones, sub-dvded by areas of populaton densty and has set mprovement targets for each area on the bass of the prevous year s performance. Utltes that succeed n mprovng by more than the requred rate can recover the hgher costs sustaned. Conversely, companes have to pay a penalty f they fal to meet the mprovement target. Interruptons due to thrd partes are not ncluded n the calculaton. The overall performance target s to brng contnuty levels up to natonal benchmark levels based on European standards: 30 mnutes of nterruptons overall per user per year n large ctes (hgh densty); 45 mnutes n medum-szed towns (medum densty): and 60 mnutes n rural areas (low densty). Other countres have smlar regmes mposed by the regulatory authortes. The UK has a number of documents makng up the dstrbuton code. One of the most mportant s G5/4, dscussed elsewhere n ths Gude, whch regulates the connecton of harmonc loads to the pont of common couplng. Measures to encourage the mprovement of contnuty are the responsblty of the Offce of Gas and Electrcty Markets (OFGEM). Most European countres are collectng data on SAIFI (System Average Interrupton Frequency Index) and SAIDI (System Average Interrupton Duraton Index): n ther formulas the number of customers s used as a bass for weghtng, but Austra, France and Span are usng the nstalled capacty (n MVA) for the same purpose, that s more consstent as weghtng factor because t corresponds to the actual electrc power that can be absorbed n each node. In MV networks the dfference of customer sze s reflected n ths case, and for operators t s easer to count capacty than the number of clents. For dstrbuton networks, SAIDI and SAIFI ndcators are used by regulators n Great Brtan, Hungary, Italy, Norway, Czech Republc, Greece, Portugal, France, Lthuana, Sweden, Estona, Ireland, Germany and the Netherlands. Although dfferent countres are usng slghtly dfferent defntons, n Europe SAIFI and SAIDI are well accepted. For transmsson networks only a few European countres are collectng the same ndces (Czech Republc, France, Portugal, Norway and Italy). Actually regulators

38 2. POWER QUALITY IN ELECTRICAL SYSTEMS prefer energy related ndcators for montorng relablty n ths case, because they generally only have a few customers who are connected drectly to ther networks. For nterruptons wth duraton shorter than three mnutes the MAIFI (Momentary Average Interrupton Frequency Index) s used. For a customer, short nterruptons are especally unpleasant n case of workng wth computers, as even an nterrupton of several seconds can lead to hgh costs. Among regulators, montorng short nterruptons s ncreasng, but stll ths phenomenon s regstered by a lmted number of countres,.e. Fnland, France, Hungary, Great Brtan and Italy, some of them on both transmsson and dstrbuton networks. Probably the man reason for ths s that for dstrbuton companes t s not easy to measure short nterruptons because they don t need human nterventon. Actually, nterruptons longer than 3 mnutes can be reported manually as number of outages, whereas for short nterruptons automatc montorng equpment should be nstalled. By comparng the data on SAIFI of the countres wthn Europe to establsh ther relablty performance, t shows that countres lke Portugal and Fnland have hgher SAIFI functons than the other countres. Anyway countres lke UK, Ireland and Netherlands have a hgher average duraton per outage. Based on the SAIDI, the Netherlands has the hghest relablty and Portugal the lowest one. Although the SAIFI s dfferent between UK and Italy, the relablty based on the SAIDI s equal for both of them. In most countres outages caused by faults n other networks are consdered as beng caused outsde of area of responsblty of the network operator. Because of ts nature, the so-called Force Majeure s more dffcult to assess, but should not be blamed to the operator. On the other hand, not ncludng such exceptonal and severe crcumstances n qualty regulaton could lead to serous debts or even bankrupt of the network operator. Therefore, these events could be better montored separately. Outages caused by thrd partes,.e. externally caused outages, and nternally caused outages are not always easy to dstngush. Moreover, externally caused outages could be nfluenced by the network operator nto some extent. For example, the operator could protect hs system better to external faults or could provde better nformaton to partes that potentally hurt the system. Moreover, the duraton of the outage can be nfluenced largely by the network operator because he s restorng power supply. Clmate and weather nfluence the qualty of supply, especally n case of overhead networks. The clmate s a factor that could not be changed by the network operator, however he s the only party who s able to optmze expendture and the fnal qualty. Tradtonally, the trade-off between qualty of supply and network cost s dfferent for rural area and urban areas. Ths results n more meshed underground networks n urban areas, and hence to a better qualty of supply. In qualty of supply regulaton often urban and

39 2. POWER QUALITY IN ELECTRICAL SYSTEMS rural networks are treated dfferently. Some European countres collect SAIDI and SAIFI data separately for dfferent customer denstes as well. Italy and Lthuana use the number of nhabtants of muncpaltes as the characterstc for classfcaton, whereas Span, Portugal and Latva use a classfcaton based on customers nstead of nhabtants. Wthn the UK each company has to report about ther avalablty. Advanced qualty regulaton should take nto account both the energy consumed by the customer and the vulnerablty of the customer. Of course, ths could not be done for every ndvdual customer connected to the dstrbuton network, however a compromse should be create customer groups and collect qualty data separately for dfferent groups. Optmal qualty s acheved f the addtonal costs to provde hgher qualty are equal to the resultng decrease n nterrupton costs experenced by the consumers. If qualty s hgher than the optmum, there s a welfare loss as consumers would be provded a level of qualty where the addtonal costs of provdng ths hgh qualty exceed the assocated reducton n nterrupton costs. Conversely, f qualty s below the optmum, there s also a reducton n nterrupton costs. The cost of an nterrupton s drven by a number of factors, frst of all ts duraton. For the ndustral sector t has been found that the cost per hour of nterrupton decrease wth duraton, suggestng that there s a large ntal fxed cost component and a varable component that decreases wth duraton. Another factor nfluencng the cost of an nterrupton s the relablty level at whch the customer s beng suppled. Generally, the hgher the relablty level the more severe the mpact of an nterrupton wll be. As the frequency of nterruptons ncreases, consumers can make a better trade-off between expected nterrupton costs and the adaptve response costs thus mnmzng total nterrupton costs. Interrupton costs vary also wth the tme of the year, day of the week and tme of the day. The regulatory cap control framework provdes companes wth strong ncentves to avod over-nvestments, reduce costs and to mprove effcency. Ths may have strong mplcatons on the short- and long-term relablty of the system. Therefore, regulators wll need to accompany prce regulaton to protect customers aganst a decrease n qualty and performance standards below certan lmts. The ntroducton of the qualty of supply regulaton s n lne wth the man task of a regulator, the protecton of customers from monopoly power of the network operators. In dong so, qualty regulaton helps to overcome ncentves to reduce qualty that are provded wthn the system of cap regulaton. Thus, qualty regulaton s a necessary component of prce regulaton to balance the ncentves to cut cost n order to provde the amount of qualty the customers expect and are wllng to pay for. Even though the qualty reducton may cause addtonal cost for network users, the monopoly network operator may stll fnd t more proftable to

40 2. POWER QUALITY IN ELECTRICAL SYSTEMS cut costs at the expense of qualty. The qualty o supply s just as mportant as prces to customers. If standard servces fall but prces reman the same, consumers are effectvely sufferng from an ncrease n prces. Another beneft of qualty regulaton s that t provdes better gudance to the regulated companes n developng and mplementng ther qualty polcy. Even f provdng hgh qualty s mportant to the network servce provders, ths does not answer the queston of how hgh ths qualty should actually be Relablty regulaton Relablty s a measure for the ablty of the network to contnuously meet the demand from customers [25]. For ts regulaton three methods can be dstngushed: a) Performance publcaton ndrect method; b) Standards; c) Incentve schemes. a) performance publcaton s wdely used by regulators. The regulator requres the companes to dsclose nformaton about trends n ts qualty performance to the publc. Overvews of the company s qualty performance are then provded, for example, n the company s annual reports, n dedcated regulatory publcaton or on the company s webste. Addtonally, the regulator can oblge the regulated servce provder to take nto consderaton the vews of customer representaton groups or nclude customers n the advsory or supervsory boards. Performance publcaton s relatvely smple to mplement and requres lmted regulatory nvolvement. The basc dea s to expose the company to publc scrutny by provdng customers wth nformaton about the company s performance. The assumpton s that the company would then be nclned to match ts qualty to customer demand because of ts reputaton. b) standards put a floor to the performance level of the company. Volaton of the standard can lead to a fne or tarff rebate. Examples of such standards are customer mnutes lost, percentage of customers wth outage, or some aggregated qualty ndex. In regulaton there s a dstncton between overall standards and guaranteed standards. Overall ones are levels of performance set by the regulator and companes must do ther utmost to comply wth them. They are not measured wth respect to performance for ndvdual customers. Guaranteed ones are levels of performance whch must be acheved n each ndvdual delvery of a specfed servce. Customers who fal to receve the requred level of servce under a guaranteed standard may be enttled to receve a penalty payment. Standards can be defned per regon or zone. In ths case, the standard s called a zonal standard. Usually, zones wth hgher customer densty, such as urban areas, have a hgher standard to reflect the hgher costs nvolved n supplyng customers lvng n rural and less densely populated zones. Consequently, the mnmum standard for urban zones

41 2. POWER QUALITY IN ELECTRICAL SYSTEMS would be set hgher than for rural ones. Wthn the regulatory practce standards tend more and more to be set as guaranteed standards snce they are easer to measure and to document. Guaranteed standards ntend to protect the sngle customer and do not ncentve the average performance of the regulated network operator. Standards are used to set lmts for commercal qualty and relablty. The man problem of a standard s that t mposes a dscrete and not contnuous relatonshp between qualty and prce. The company ether pays a fne or t does not, dependng on whether t volates the set standard: there s nothng n between. The queston s at what level the standard should be set, and what the level of the fne should be. These two need to be low enough to be defensble and hgh enough to be effectve. If they are set too hgh the standard may severely punsh the company for not meetng unrealstc targets. If the standard s set too low, qualty degradaton may occur. Qualty ncentve schemes can be consdered as an extenson of a standard. Alternatvely, a standard can be consdered as a specal case of qualty ncentve scheme. The prce and qualty are closely related and the company s performance s compared to some qualty target: devatons result n ether a penalty or a reward there are many varatons of qualty ncentve schemes. Prce and qualty can be mapped contnuously, n a dscrete way, or a combnaton of these; the level of the penalty can be capped, dead bands can be appled. Fgure 2-4 shows some examples, where the x-axs represents the measured qualty level, the y-axs the penalty or reward. Fgure 2-4. Penaltes/Rewards n functon of qualty. Qualty ncentve schemes can be used for all knd of qualty ndcators. The measured performance can be expressed for example n terms of SAIDI or SAIFI. The Dutch regulator has ntroduced a qualty ncentve scheme that refers to these ndces. Fgure 2-5 reports the Dutch ncentve scheme

42 2. POWER QUALITY IN ELECTRICAL SYSTEMS Fgure 2-5. Qualty ncentves scheme for SAIDI and SAIFI. Besde the three methods of qualty regulaton that are normally appled together regulatory practce s facng more and more tendency towards so-called ntegrated prcequalty regulaton. It solves the trade off between cost and qualty by explctly consderng qualty as a cost component wthn the benchmarkng snce t can be assumed that hgher qualty leads to hgher cost and vce versa. By dong so, qualty s taken nto account whle comparng the effcency of the network operators wthn the analyss of data wth a method to compare frms usng multple nput and output factors. Up to ths pont only short tme measurement and short tme mplcatons of prce regulaton on relablty have been consdered. Anyway, gven the long term nature of nvestment decsons and the effects of a contnuous mantenance, short term decsons on qualty have a deep mpact on future cost and qualty that cannot be under control just wth short tme measurements. Long term analyss and assessment of relablty s becomng more and more mportant. Regulators should be aware of the nteracton between short term ncentves and long term consequences of ther decson and use of addtonal tools to evaluate them. Long tme relablty control should be n lne wth the general regulatory approach. It comes clear that cap regulaton provdes strong ncentves to reduce cost more than oblged by the regulator n order to realze effcency gans, leadng to an overall decrease n qualty of supply and qualty restrcton for certan customer groups. Qualty regulaton s mportant as a part of ncentve regulaton n order to ensure approprate solutons for the cost-qualty trade off respectng the customers demand for relablty. By applyng the above three methods the cost-qualty trade off s not mmedately solved. Hence, regulators ntroduce ntegrated prce-qualty regulaton that consders the qualty provded n the effcency analyss. Moreover, the long-term aspects of qualty become more and more a challenge for regulators to mplement a balanced system that ensures the consstency of short tme effcency ncentves and long term relablty

43 2. POWER QUALITY IN ELECTRICAL SYSTEMS Voltage qualty regulaton In many countres voltage qualty s regulated to some extent, often usng ndustral standards or accepted practce to provde ndcatve levels of relevant performance. The man dfference n voltage qualty and network relablty s that untl a certan voltage qualty level the customer s not affected by a not perfect performance, whereas the same customer s affected by any nterrupton of the power supply. The customer does not have nterest n mprovng the voltage qualty as long as t stands wthn certan lmts, whle the customer has nterest n avodng all nterruptons n the power supply. In most European countres the voltage qualty s not an ssue for a large majorty of the customers n dstrbuton networks. Ths means that these customers wll bascally not beneft from mprovements n the voltage qualty. However, as connected equpment s not workng due to the lack of voltage qualty, mprovement of the qualty has a value to the customer. Because some equpment s more vulnerable to lack of techncal qualty than other equpment, some clents wll value ths qualty ncrease dfferently. Therefore, t s hard to assess the value of techncal qualty and ts ndvdual dmensons than the value of prevented nterruptons. Ths s the second man dfference between voltage qualty and network relablty. A thrd dfference s the cause of the lack of qualty. Whle power nterruptons for the customer are manly caused n the electrcty network or the connecton of the customer, voltage qualty s largely nfluenced by other customers. Harmonc dstorton s caused by electronc equpment connected to the network and voltage dps could be caused by short crcuts n the network or by weldng apparatus. Hence the network operator should not pay for every sngle decrease of voltage qualty, but t should be able to keep the voltage qualty at least on acceptable levels by applyng maxmum dsturbance levels and checkng customers on keepng them. In most countres mnmum standards are defned for voltage qualty. Ths way f a mnmum level of qualty s met, the customer s not nterested n a better qualty, anyway the network operator s responsble for meetng a mnmum of qualty n hs system. Snce there s a large dfference regardng the nfluence of lack of qualty for dfferent type of customers, some customers need hgher qualty than the mnmum standard levels. In most cases, ths s regulated by connecton contracts. Dutch network operators are nvestgatng a transparent classfcaton system for the delvered voltage qualty on the pont of common couplng. Actually, the mnmum standards are usually based on an nternatonal accepted standards, as the European EN the characterstc of the supply voltage concernng: frequency, magntude, waveform and symmetry of the phases. Mostly these mnmum standards for voltage qualty are ncluded n the grd code or the dstrbuton code. Although EN gves ndcatve values for many of the phenomena, t s only applcable to voltage levels up to

44 2. POWER QUALITY IN ELECTRICAL SYSTEMS 35 kv. For hgher voltage levels no standard exsts; n the Netherland, Italy and Portugal some crtera from EN are extended to voltage levels up to 50 kv or hgher. In some countres, voltage qualty standards ntroduced by regulators dffer from the lmts prescrbed by EN 50160; n an ncreasng number of EU countres the EN reference levels are not found to be satsfactory both by regulators and customers. The CEER Benchmarkng report and the Cgré workgroup on PQ both hghly recommend that EN should be revsed, takng nto account both the actual levels of voltage qualty n European transmsson and dstrbuton networks, the evaluaton of customer s needs. Lke network relablty, the voltage qualty n dstrbuton networks s nfluenced as well by the voltage qualty n the nterconnected transmsson and dstrbuton networks. A dfference however s that where nterruptons n the power supply could be related to a cause n a network, voltage qualty s the result of many dfferent causes whch are changed durng transmsson and transformaton. Thus, t s hard to dentfy the network owner who should be responsble for the level of power qualty Montorng the voltage qualty wthn EU A good knowledge of the real stuaton s a prelmnary step towards any knd of regulatory nterventon. Therefore, a growng number of European countres have montorng systems nstalled or plan to nstall them n the near future, as shown n Table 2-5. Montorng systems wthn a number of EU countres are based on a samplng ether of transmsson-dstrbuton nterface ponts or customer connecton ponts. Table 2-5. Montorng system n European Countres. VOLTAGE QUALITY MONITORING SYSTEM Montorng at both transmsson and dstrbuton levels Montorng only at transmsson level Montorng only at dstrbuton level Proposal stage Countres Italy, Norway, Portugal, Slovena, the Netherlands Czech republc Hungary Span and Sweden In Norway a montorng system has been appled for several years. From 2006 mandatory voltage qualty montorng started: each network company s oblged to montor qualty parameters contnuously n dfferent characterstc parts of ts power system. In Hungary the regulator owns 400 voltage qualty recorders that are nstalled each semester n one of the sx dstrbuton companes, at low voltage only. The regulator chooses the network ponts randomly, n a way that does not depend on prevous events or complans

45 2. POWER QUALITY IN ELECTRICAL SYSTEMS In Portugal there are 61 ponts montored on the transmsson grd (40 for 4 weeks and the rest all year long); n dstrbuton system, all 423 substatons n MV and 1270 power transformaton statons n LV have been montored for 3 years. In Slovena dstrbuton and transmsson companes are oblged to measure voltage qualty parameters; montorng s mplemented n hgh voltage coverng all the substatons and about 10% of MV systems. In Italy at the end of 2004 the regulator asked the transmsson company to nstall about one hundred voltage qualty recorders; as for dstrbuton, a voltage qualty montorng system of 400 ponts s workng n about 10% of MV bus-bars n HV/MV transformers. In Span the dstrbuton companes and the regulator have been workng on a procedure for controllng and measurng voltage qualty; 10% of the busbars n MV of each provnce s nvolved. In Czech Republc a montorng systems s gong to be nstalled at the nterconnecton ponts between transmsson and dstrbuton networks. In the Netherlands the grd operators measure at 150 ponts (50 ponts at HV, 50 ponts at MV and 50 ponts at LV) the qualty for one week each. Every year these ponts are selected randomly n such a way that does not depend on prevous events or complans. The measurng devces are owned by the federaton of energy companes n the Netherlands. Wth a lmted number of measurng equpment 150 network ponts can be montored. The grd operators started n 2005 wth measurng the voltage dps at 20 EHV statons and 20 HV statons for a perod of one year. Although these montorng systems wthn several countres are dfferent from each other n many respects, a common pont s that at least voltage magntude, dps and harmonc dstorton of the voltage waveform are montored. The number and locaton of voltage recorders s qute dfferent from one country to another. As for ndvdual voltage qualty measurements, one case deserves specal attenton. In France the man dstrbuton companes offer ther customers customzed contracts wth assgned power qualty levels. If the customer clams for better contractual levels than normal ones, he can ask the operator for customzed contractual levels n s contract, payng an extra charge. Customers havng customzed contracts must be montored by a recorder nstalled and owned by the customers themselves or by the operator. In dstrbuton networks about 16% of MV customers have a power qualty recorder, whereas n the transmsson networks the montorng nvolves about 12% of EHV and HV customers. In comparson to regulaton of relablty, the regulaton of voltage qualty s less advanced. Where some European countres are usng complex and hghly effectve regulatons (e.g

46 2. POWER QUALITY IN ELECTRICAL SYSTEMS ncentve regulaton) for relablty, for regulaton of voltage qualty most regulators rely on ndrect measurements or mnmum standards at most. As defntons and procedures are n place, the relablty measurement s qute mmedate, no specal measurement equpment needs to be nstalled to collect duraton of the power nterrupton and the number of affected customers, or the total amount of nterrupted power. Ths s dfferent for most dmensons of voltage qualty, whch need to be measured wth specalzed measurement devces. An mportant ssue s that the voltage qualty s dfferent for every connecton pont n the network. Because t s not feasble to measure everywhere, statstcal technques are needed to report on average voltage qualty of ndvdual senstve customers. In contradcton to ths, relablty can be measured by just sortng out and summng all ndvdual outage statstcs. One more dfference between relablty and power qualty s the cause of ther lack. For nterruptons n most cases the cause could be found n one or more events n the publc electrcty network, whereas voltage qualty s nfluenced by both condtons n the publc electrcty network and at the clent ste. Steel manufacturers are well known producers of voltage dps that are exported from ther plants to the network, nfluencng the voltage qualty of other clents. Although the network operator should be responsble for the voltage qualty n ts network, ths aspect makes the voltage qualty regulaton a more complex ssue. In a number of EU countres (Italy, Norway, Portugal, Slovena, Czech Republc, Hungary and the Netherlands) voltage qualty montorng systems are nstalled or currently under commssonng. Most countres apply recorders on both transmsson and dstrbuton, whle Hungary, e.g., s nterested n qualty only of low voltage. Anyway, n most countres there s no systematc montorng system for voltage qualty, but a clear trend shows that the number of countres wth montorng systems s ncreasng. For the measurement of system qualty, n Norway, Hungary, Portugal, Slovena, Netherlands, Italy and Span a measurement program s or s planned to be nstalled. The program statstcally determnes parameters that provde general pcture of voltage qualty of the system by usng a large set of voltage qualty meters. For example, n Hungary 400 devces are nstalled n the low voltage network, for one semester at each place. In Portugal 1270 low voltage substatons 423 medum voltage substatons and 61 ponts n the transmsson grd are montored. Although measurement schemes are nstalled more and more, results of voltage qualty are not yet easly accessble for the average clent Mnmum standards Snce voltage qualty s an ndvdual ndcator, n almost all European countres a set of mnmum standards are ntroduced to defne the mnmum voltage qualty for ndvdual connecton ponts to be delvered by the network operator. Many EU countres apply EN as the mnmum standard for voltage qualty even f ths standard s recognzed to

47 2. POWER QUALITY IN ELECTRICAL SYSTEMS be not perfect. One mportant dsadvantage s that EN s vald only for voltage levels up to 35 kv, but n many countres the same or smlar levels for hgher voltage networks are used. Moreover, EN only provdes mandatory standards for a lmted number of Power Qualty ndcators, whle for others only ndcatve values are provded. In addton to ths, many standards are defned for 95% of the tme, leavng out the rest of the tme. Fnally, sometmes EN s consdered to be too weak. Although EN s very common for European Regulators, a number of them s adaptng ther mnmum standards on voltage qualty so that the relevant dsadvantages are overcome. E.g. Norway has adapted voltage qualty standards on supply voltage varatons, flcker severty, rapd voltage changes, voltage unbalance and harmonc dstorton. The Norwegan standards are now made better compatble to mmunty levels of equpment. Also France, the Netherlands and Portugal have adapted some of the EN standards. Currently several nsttutons such as Cgré and Cred are dscussng mprovements to the standard EN If one ssue s havng sensble standards, another one s how to deal wth voltage qualty that does not meet the mnmum standards. A large number of countres (Austra, Belgum, Czech Republc, Estona, France, Latva, Norway and Poland) dstrbuton network operators have the oblgaton to verfy voltage qualty complants of ndvdual customers. Generally, ths s done on customer s expenses, but sometmes customers only pay f the voltage qualty meets the standards. Usually, not meetng the voltage qualty standards does not lead to penaltes and only leads to the oblgaton for the network operator to mprove qualty n order to fulfll the mnmum standards. Some countres have nstalled a complant procedure, whch ncludes a maxmum response tme for the network operators on power qualty complants of the customer. Sometmes a penalty payment s needed f the network operator exceeds ths set tme. In the U.K. n case of voltage complant by a customer, the dstrbuton network operator must vst the customer wthn 7 workng days or send a substantve reply wthn 5 days. If the operator fals meetng ths standard, the customer s provded wth a payment of 20. smlar complant procedures exst n Norway, Hungary, Ireland, Italy, Latva, Portugal and Slovena. Penaltes appled n these countres range from 8 Euro for 7 workng days n Hungary (n case of domestc customers) to 75 Euro for 15 workng days n Portugal (Medum and Hgh voltage customers). Hungary and Ireland apply also a standard for the correcton of voltage qualty problems. In the frst country the operators need to pay a penalty of 20 to 120 Euro (dependng on the sze of the customer) f voltage complants are not compensated wthn 12 months. In the second country a payment of 50 Euro s needed after three months

48 2. POWER QUALITY IN ELECTRICAL SYSTEMS Incentve schemes Unlke power qualty regulaton wth regard to relablty, currently no regulators are applyng ncentve regulaton for voltage qualty. However, some countres are applyng so-called power qualty contracts,.e. ndvdual contracts between network operator and customer agree on voltage qualty standards whch are dfferent from the usual standard. In France network operators usually offer all customers such contracts, whch could be customzed to the desred qualty level. The payment s related to the work that has to be done by the operator n order to meet these standards. Whle ths contract s pretty popular for relablty (for whch 1000 over 100,000 MV customers have a contract) only 92 customers have a contract wth customzed contractual levels on voltage qualty. Because of dfferent reasons, ncentve regulaton schemes are not yet n place for qualty regulaton. Snce nterruptons are consdered to be more mportant by the majorty of customers, regulators started wth ncentve regulaton for relablty. On the other hand, some nvestgatons show that costs of lack of qualty are sgnfcant, e.g. Norwegan nvestgaton shows that customers costs assocated wth short nterruptons and voltage dps n Norway are smlar to the customers costs related to long nterruptons. However, some ssues need to be addressed before beng able to mplement ncentve regulaton. Man ssues are: Measurement of short nterruptons and long and deep voltage dps; Determnng customers costs n case of short nterruptons and long and deep voltage dps. Table 2-6. Methods for qualty regulaton n Europe. Method APPLIED FOR VOLTAGE QUALITY Objectve Indrect Standards Incentve scheme Montorng voltage qualty by large measurement programs Indvdual guaranteed standards based on EN Qualty contracts Not appled Montor long term development Protecton of ndvdual customer groups Meetng voltage qualty requrements of ndvdual customers Ensurng an average voltage qualty level Table 2-6 provdes a hgh-level overvew of the methods used for qualty regulaton n Europe, categorzed n ndrect methods, mnmum standards and ncentve regulaton. In some countres large voltage qualty montorng systems are nstalled. Usng results relevant to more than one year the long term developments of voltage qualty of the system could be montored. Snce equpment at the customers wll get more senstve for lack of voltage qualty t can be expected that the regulators wll extend regulaton on voltage qualty. A better nsght

49 2. POWER QUALITY IN ELECTRICAL SYSTEMS n costs at customers due to lack of voltage qualty, characterstc of regulaton, ndvdually for the dfferent parameters of voltage qualty could be helpful for them. It s mportant to consder all voltage qualty parameters ndvdually, e.g. regulatng on voltage dps wll be dfferent from regulatng on harmoncs

50 2. POWER QUALITY IN ELECTRICAL SYSTEMS References [1] EN 50160, Voltage characterstcs of electrcty suppled by publc dstrbuton systems, CENELEC, Bruxelles (Belgum), 1999; [2] IEEE std , Recommended practce for montorng electrc power qualty, The IEEE, Pscataway (USA), Nov. 1995; [3] IEEE Tral-Use Std , Defntons for the measurements of electrc power quanttes under snusodal, nonsnusodal, balanced or unbalanced condtons, The IEEE, New York (USA), June 2000; [4] IEEE Tral-Use Std , Recommended Practces and Requrements for Harmonc Control n Electrcal Power Systems The IEEE, Pscataway (USA), June 1992; [5] IEEE Std , IEEE Recommended Practce for the Desgn of Relable Industral and Commercal Power Systems, The IEEE, Pscataway (USA), 2007; [6] IEEE Std , Recommended Practce for Evaluatng Electrc Power System Compatblty wth Electronc Process Equpment, The IEEE, Pscataway (USA), May 1998; [7] IEC X, Electromagnetc compatblty (EMC): Envronment, 2002; [8] IEC , Electromagnetc compatblty (EMC): Envronment Voltage dp and short nterrupton on publc electrc power supply system wth statstcal measurements results ; [9] IEC X, Electromagnetc compatblty (EMC): Lmts ; [10] IEC , Electromagnetc compatblty (EMC): Lmts for harmonc current emssons (equpment nput current 16 A per phase) ; [11] IEC , Electromagnetc compatblty (EMC) Part 3:Lmts Assessment of emsson lmts for dstortng load n MV and HV power systems ; [12] IEC X, Electromagnetc compatblty (EMC): Testng and Measurements technques, 2002; [13] IEC , Electromagnetc compatblty (EMC) Part 4-7: Test and Measurements technques General gude on harmoncs and nter-harmoncs measurements and nstrumentaton for power supply systems and equpment connected thereto ; [14] IEC , Electromagnetc compatblty (EMC) Part 4-15: Test and Measurements technques Flckermeter Functonal and desgn specfcatons ; [15] IEC , Electromagnetc compatblty (EMC) Part 4-30: Test and Measurements technques Power qualty measurements methods ; [16] Autortà per l Energa Elettrca e l Gas: Testo ntegrato delle dsposzon dell Autortà n matera d qualtà de servz d dstrbuzone, msura e vendta dell energa elettrca, Delbera 30 gennao 2004, n.4/04; [17] E. J. Davs, A. E. Emanuel, D. J. Plegg, Evaluaton of sngle-pont Measurements Method for Harmonc Polluton Cost Allocaton, IEEE Trans. on Power Delvery, vol. 15, n. 1, 2000, pp ; [18] P. J. Rens, P. H. Swart, On Technques for the Localzaton of Multple Dstorton Sources n three-phase Networks: Tme Doman Verfcaton, ETEP, Vol. 11, No 5, 2001, pp ; [19] C. Muscas, Assessment of Electrc Power Qualty: Indces for Identfyng Dsturbng Loads, ETEP, Vol. 8, No. 4, 1998, pp ;

51 2. POWER QUALITY IN ELECTRICAL SYSTEMS [20] L. Crstald, A. Ferrero, S. Salcone, A dstrbuted Measurement System for Electrc Power Qualty Measurement, IEEE Trans. on Instr. and Meas., Vol. 51, No. 4, 2002, pp ; [21] D. Castaldo, A. Testa, A. Ferrero, S. Salcone, An Index for Assessng the Responsblty for Injectng Perodc Dsturbances, L energa elettrca, vol. 81, 2004, Rcerche ; [22] A. Ferrero, R. Sasdell, Revenue and Harmoncs: a Dscusson about New Qualty Orented Measurement Methods, Proc. Of the 9 th ntern. Conference on meterng and tarffs for energy supply, Publcaton 462, Brmngham, UK, 1999, pp ; [23] R. Sasdell, C. Muscas, L. Peretto, A VI-based measurement system for sharng the customer and supply responsblty for harmonc dstorton, IEEE Trans. on Istr. And Meas., 1998, Vol. 47, No. 5, pp ; [24] C. Muscas, L. Peretto, S. Suls, R. Tnarell, Implementaton of mult-pont Measurement Technques for PQ Montorng, Proc. Of the 21st IEEE IMTC/04, Como(Italy), 2004, vol. 3, pp ; [25] K. Keller, B.F.C Franken, Qualty of Supply and Market Regulaton; survey wthn Europe, KEMA Consultng by order of the European Copper Insttute, Arnhem, The Netherlands, December

52

53 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3. Electromagnetc transents n power dstrbuton networks. Glossary The defntons reported n the followng have been developed and then harmonzed by several standards-wrtng organzatons. surge protectve devce (SPD) - A devce that s ntended to lmt transent overvoltages and dvert surge currents. It contans at least one nonlnear component. surge reference equalzer - A surgeprotectve devce used for connectng equpment to external systems whereby all conductors connected to the protected load are routed, physcally and electrcally, through a sngle enclosure wth a shared reference pont between the nput and output ports of each system. Sharng the references can be accomplshed ether by a drect bond or through a sutable devce mantanng solaton durng normal condtons but an effectve bondng by means of a surge-protectve devce durng the occurrence of a surge n one or both systems. back flashover (lghtnng) - A flashover of nsulaton resultng from a lghtnng stroke to part of a network or electrcal nstallaton that s normally at ground potental. blnd spot - A lmted range wthn the total doman of applcaton of a devce, generally at values nferor to the maxmum ratng. Operaton of the equpment or the protectve devce tself mght fal n that lmted range despte the devce s demonstraton of satsfactory performance at maxmum ratngs. drect stroke - A stroke mpactng the structure of nterest or the sol (or objects) wthn a few metres from the structure of nterest. energy deposton - The tme ntegral of the power dsspated n a clampng-type surge protectve devce durng a current surge of a specfed waveform. falure mode - The process and consequences of devce falure. faclty - Somethng (as a hosptal, machnery) that s bult, constructed, nstalled or establshed to perform some partcular functon or to serve or facltate some partcular end. follow current - Current suppled by the electrcal power system and flowng through the SPD after a dscharge current mpulse and sgnfcantly dfferent from the contnuous operatng current. leakage current - Any current, ncludng capacty coupled currents, that can be conveyed from accessble parts of a product to ground or to other accessble parts of the product. lghtnng protecton system (LPS) The complete system used to protect a space aganst the effects of lghtnng. It conssts of both external and nternal lghtnng protecton systems. In partcular cases, an LPS mght consst of an external LPS or an nternal LPS only. lghtnng flash to earth - An electrcal dscharge of atmospherc orgn between cloud and earth consstng of one or more strokes. lghtnng stroke - A sngle electrcal dscharge n a lghtnng flash to earth. mans - The a.c. power source avalable at the pont of use n a faclty. It conssts of the set of electrcal conductors (referred to by terms ncludng servce entrance, feeder, or branch crcut ) for delverng power to connected loads at the utlzaton voltage level

54 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS measured lmtng voltage - The maxmum magntude of voltage that appears across the termnals of the SPD durng the applcaton of an mpulse of specfed wave-shape and ampltude. nearby strke - A strke occurrng n the vcnty of the structure of nterest. open-crcut voltage (OCV) - The voltage avalable from the test set up (surge generator, couplng crcut, back flter, connectng leads) at the termnals where the SPD under test wll be connected. pont of strke - The pont where a lghtnng stroke contacts the earth, a structure, or an LPS. pulse lfe - The number of surges of specfed voltage, current ampltudes, and wave-shapes that may be appled to a devce wthout causng degradaton beyond specfed lmts. The pulse lfe apples to a devce connected to an a.c. lne of specfed characterstcs and for pulses suffcently spaced n tme to preclude the effects of cumulatve heatng. short-crcut current (SCC) - The current whch the test setup (surge generator, couplng crcut, back flter, connectng leads) can delver at the termnals where the SPD under test wll be connected, wth the SPD replaced by bondng the two lead termnals. standby current - The current flowng n any specfc conductor when the SPD s connected as ntended to the energzed power system at the rated frequency wth no connected load. surge response voltage - The voltage profle appearng at the output termnals of a protectve devce and appled to downstream loads, durng and after a specfed mpngng surge, untl normal stable condtons are reached. swell - A momentary ncrease n the power frequency voltage delvered by the mans, outsde of the normal tolerances, wth a duraton of more than one cycle and less than a few seconds. temporary overvoltage - An oscllatory overvoltage (at power frequency) at a gven locaton, of relatvely long duraton and whch s not damped or weakly damped. Temporary overvoltages usually orgnate from swtchng operatons or faults (e.g.; sudden load rejecton, sngle-phase faults) or from non-lnear phenomena (ferro-resonance effects, harmoncs). thermal runaway - An operatonal condton when the sustaned power loss of an SPD exceeds the dsspaton capablty of the housng and connectons, leadng to a cumulatve ncrease n the temperature of the nternal elements culmnatng n falure. two-port SPD - An SPD wth two sets of termnals, nput and output. A specfc seres mpedance s nserted between these termnals. The measured lmtng voltage mght be hgher at the nput termnals than at the output termnals. Therefore, equpment to be protected must be connected to the output termnals. voltage-lmtng type SPD - An SPD that has a hgh mpedance when no surge s present, but wll reduce t contnuously wth ncreased surge current and voltage. Common examples of components used as nonlnear devces are: varstor and suppressor dodes. These SPDs are sometmes called clampng type. voltage-swtchng type SPD - An SPD that has a hgh mpedance when no surge s present, but can have a sudden change n mpedance to a low value n response to a voltage surge. Common examples of components used as nonlnear devces are: spark-gaps, gas tubes, thyrstors and tracs. These SPDs are sometmes called crowbar type

55 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3.1. Transents Fgure 3-1. Voltage varatons represented accordng to ther duraton, magntude and effect on the power system Transent overvoltages and currents, globally referred to as surges, occur n power systems as the result of several types of events or mechansms, and can be classfed n four categores: 1. Lghtnng surges; They are the result of drect strokes to the power or communcatons systems, or surges caused by lghtnng strkng structures (wth or wthout lghtnngprotecton system) or the sol. Lghtnng surges can mpact a faclty by mpngng upon the servce entrance as they propagate along the conductors, havng orgnated from one of three phenomena: a drect stroke to the lnes; a resstvely-coupled surge from a nearby strke; a voltage nduced n the loops formed by lne conductors and earth. Lghtnng surges can also mpact a faclty as they are coupled drectly nto the faclty wrng system, wthout havng been brought to the servce entrance, as descrbed above. Several mechansms are nvolved n ths process: the earthseekng, fast-changng current of a drect stroke to the faclty - ether to ts ntended lghtnng protecton system (LPS) or to other structures (roof mounted equpment n partcular) - wll nduce transent voltages n the faclty crcuts; the slower porton of the lghtnng current wll couple transent voltages nto the faclty wrng through common mpedances; the earth-seekng current

56 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS flowng n the faclty wrng wll dvde among the paths avalable for dsperson,.e. earthng electrode(s) of the faclty, and any utlty metallc path enterng the faclty. 2. Swtchng surges; Swtchng surges are the result of ntentonal actons on the power system, such as load or capactor swtchng n the transmsson or dstrbuton systems by the utlty, or n the low-voltage system by end-user operatons. They can also be the result of unntentonal events such as power system faults and ther elmnaton. Transents assocated wth swtchng surges nclude both hgh-frequency transents and low-frequency transents. Hgh-frequency transents are generally assocated wth natural oscllatons of the crcut elements n response to the stmulus of a change of state n the crcut. They nvolve relatvely small stray capactances and nherent nductances, hence ther hgh frequency, and relatvely low energy-delvery capablty n a drect mpact to the power port of equpment. On the other hand, the hgh frequency has the potental of couplng nterference, rather than cause damage. In partcular, fast swtchng currents n the power system can nduce nterferng voltages nto control crcutry n the faclty. Ths scenaro mght appear outsde of the scope of power qualty but should stll be consdered because ts effects mght mmc a power qualty problem mpngng the power port of equpment. Low-frequency transents are prmarly assocated wth the swtchng of capactor banks. These banks can be a part of the utlty system or of the faclty. In general, they nvolve kvar or MVAr, and therefore have consderable energy-delvery capablty nto an SPD that would attempt to dvert them, or nto the ntermedate d.c. lnks of electronc power-condtonng equpment. Because of ther low frequency, they can propagate a consderable dstance from the pont of orgn because the wrng nductance does not have the mtgaton effect avalable for the hgher-frequency transents. Therefore, these capactor-swtchng transents mert partcular attenton. 3. Temporary overvoltages; They occur n power systems as the result of a wde range of system condtons, both normal operaton and abnormal condtons. These overvoltages mert attenton because they not only cannot be mtgated by SPDs - the normal response of a desgner confronted wth transents - but also represent a sgnfcant threat to the survval of SPDs. For ths reason,

57 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS whle common wsdom mght be that ther power-frequency nature places them outsde the scope of transents, relable applcaton of SPDs demands that they be taken nto consderaton. 4. System-nteracton surges. Transent overvoltages can occur between dfferent systems, such as power and communcatons, durng the flow of surge currents n one of the systems. These nteractons mght be deemed outsde of the scope of Power Qualty at the power port of equpment, but ther effect can gve the appearance of a power qualty problem, and therefore they mert recognton f a soluton s to be found. A system-nteracton overvoltage s partcularly mportant for equpment that s connected to both the power system and some form of communcatons systems, whch s ncreasngly the case of ndustral equpment. Because the two systems can be managed by dfferent ndvduals or organzatons, even though nstalled n the same faclty, the earthng practces appled by these separate groups are often uncoordnated at best and counter-productve at worst. Broadly speakng, the transents can be classfed nto two categores mpulsve and oscllatory. These terms reflect the waveshape of a voltage or current transent. Standard IEEE 1159 [1] takes nto account also the frequency content of the event assocated wth ts rsng front rate and ts duraton. Table 3-1 reports such classfcaton. Transents Categores Impulsve Table 3-1. Voltage transent phenomena. Typcal spectral content Typcal duraton Nanosecond 5 ns rse < 50 ns Mcrosecond 1 µs rse 50 ns 1 ms Mllsecond 0.1 ms rse > 1 ms Oscllatory Typcal voltage magntude Low frequency < 5 khz ms 0-4 pu Medum frequency khz 20 µs 0-8 pu Hgh frequency MHz 5 µs 0-4 pu Lghtnng The sgnfcant lghtnng parameters nclude waveforms, ampltudes, and frequency of occurrence. The lterature contans data obtaned by measurements as well as data produced by computatons. Three types of couplng mechansms can produce overvoltages n low-voltage systems. Whle ths dscusson makes reference to overvoltages, consderaton of the current assocated wth the overvoltage, or the

58 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS current ntally causng the overvoltage, s an mportant aspect of the subject. In the case of a drect stroke to the electrcal system, the mmedate threat s the flow of lghtnng current through the earthng mpedances, resultng n overvoltages. The effectve mpedance of the lghtnng channel s hgh (a few thousand ohms). Accordngly the lghtnng current can practcally be consdered an deal current source. In case of a near flash, the mmedate threat s the voltage nduced n crcut loops, whch n turn can produce surge currents. In the case of a far flash, the threat s lmted to nduced voltages. Therefore, the response of an electrcal system to the lghtnng event s an mportant consderaton n assessng the threat. For a gven flash, the severty of the overvoltage appearng at the end-user faclty reflects the characterstcs of the couplng path, such as dstance and nature of the system between the pont of flash and the end-user faclty, earthng practces and earth connecton mpedance, presence of SPDs along the path, and branchng out of the dstrbuton system. All of these factors vary over a wde range accordng to the general practce of the utlty as well as local confguratons. The world-wde annual frequency of thunderstorm days s shown n Fgure 3-2. Long used for rsk assessment, ths nformaton s now beng superseded by maps of flash densty for regons where a lghtnng detecton system s n operaton. Flash densty maps provde more accurate nformaton than the tradtonal thunderstorm day, and they are expected to supersede the thunderstorm maps as they become avalable. Lghtnng surges n electrcal systems can n general be classfed accordng to ther orgn as follows: Current surges due to drect flashes to overhead lnes, ncludng back flashover events; Induced overvoltages on overhead lnes due to flashes at some dstance, and the resultng surge currents; Overvoltages caused by resstve, nductve, and capactve couplng from systems carryng lghtnng currents, and the resultng surge currents. In the followng paragraphs, these classes of overvoltages are brefly descrbed Drect flashes to overhead lnes The effectve mpedance of the lghtnng channel s hgh and the lghtnng current can practcally be consdered as an deal current source. The resultng overvoltages are therefore determned by the effectve mpedance that s seen by the lghtnng current. For a flash to an overhead lne conductor, the mpedance n the frst moments s determned by the characterstc surge mpedance of the lne. Gven the typcal values of characterstc mpedances, rangng from tens of ohms to 400 ohms, very hgh overvoltages occur that can be expected to cause flashover to earth long before the servce entrance of a buldng becomes nvolved. Therefore, the lghtnng surge appearng at servce entrances, whle reflectng the severty of the lghtnng stroke and ts dstance, bears no resemblance to the actual lghtnng current

59 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Fgure 3-2. Example of world-wde annual frequency of thunderstorm days Induced overvoltages on overhead lnes Due to the changes n electromagnetc feld caused by a lghtnng flash, surges are nduced n overhead lnes of all knds, also at consderable dstance from the flash. The voltages have essentally the same value for all conductors because the phase separaton s small compared to the dstance to the flash. For nstance, n a hgh-voltage lne wth 10-m conductor heght for a lghtnng current of 30 ka, the nduced voltage s n the order of 100 kv for a flash at 100-m dstance. For a low-voltage lne wth a heght of 5 m, a current of 100 ka wll nduce a voltage of about 2 kv even at a dstance of 10 km. However the nduced voltage does not necessarly appear at the servce entrance: the hgh levels wll provoke flashover to earth or operaton of a surge arrester, so that the surges appearng at the servce entrances are more lkely to be n the range of only a few klovolts. As noted before, these surges nvolve sgnfcant overvoltages but relatvely small current levels, n contrast wth surges resultng from drect strkes where the currentsource lke phenomenon results n current surges that reflect the dsperson of the orgnal stroke current among the earth paths Overvoltages caused by couplng wth other systems A lghtnng flash to earth or to a part of a system normally at ground potental can result n an earth potental of hgh value at the pont of strke and n the vcnty. Ths phenomenon causes overvoltages n electrcal systems usng ths pont of earth as reference. At frst, the potental of the earth electrode s determned by the local mpedance that, for nstance, mght be 10 ohms: a hgh voltage s produced between the earthng system and electrcal

60 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS nstallatons nsde the buldng, wth a hgh probablty of causng ether nsulaton breakdown or operaton of SPDs. Followng such events, current mpulses can flow nto the varous systems, manly determned by ther mpedance to earth. In ths way overvoltages are produced n the power supply system as well as n other connected servces (telecommuncaton, data and sgnallng systems, etc.), but also transferred to other buldngs, structures, and nstallatons. For nstance, all power nstallatons suppled from the same dstrbuton transformer as the one struck by lghtnng can be affected. Due to the hgh electromagnetc felds caused by the lghtnng current, nductve and capactve couplng to electrcal systems that are close to a lghtnng path can also cause over-voltages on electronc and data systems, causng falures and malfunctons Lghtnng surges transferred from MV systems Because ther structures are longer and hgher than other structures located n ther vcnty (houses, trees), MV overhead lnes are n general more exposed to lghtnng than LV lnes. The propagaton of the surge through the MV system and the transfer rate to the LV system depend on the physcal constructon of the system. Some mportant dfferences can exst between the desgns used n dfferent countres. The lghtnng surges n MV systems are caused by drect flashes or nduced by nearby flashes. In addton, back-flashovers can occur from flashes strkng earth wres or extraneous metal parts of structures or equpment, or strkng the earth close to a lne structure Surge magntude and propagaton n MV systems The surge propagaton depends on the MV system structure and, n partcular, on the surge-protectve devces nstalled. Hgh-level lghtnng surges are n general attenuated quckly durng ther propagaton on the lne by losses and flashover across the lne nsulators. In practce, after a few lne spans, the magntude of an overvoltage s reduced to the nsulaton levels of the lne solators. Wth the excepton of drect strokes next to the MV/LV transformer, t can be assumed that overvoltages n an MV system are lmted by the nsulaton level of the lne solators. In a 20 kv system; ths s about 150 kv to 180 kv. For wood-pole lnes wthout earthed cross-arms, however, much hgher surges can occur. A second lmtaton of the surge level s provded by the surge-protectve devces located usually at the prmary sde of the MV/LV transformer, or at the entrance of an underground network. These protecton devces mght be ZnO or SC surge arresters or ar gaps. The resdual overvoltage (n the range of 70 kv for a 20 kv system, for nstance) depends on the rated value and earthng mpedance of the protecton devces. When ar gaps are used, one can expect the lghtnng surge to be followed by a power frequency follow current that can generate a temporary overvoltage

61 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Surge transfer to the LV system The overvoltage surges generated n the MV system by lghtnng are transferred to the LV dstrbuton system by capactve and magnetc couplng through the MV/LV transformer or by earth couplng. The transferred surge magntude depends on many parameters, such as LV earthng system (TT; TN, IT), LV load, LV surge-protectve devces, couplng condtons between MV and LV earthng, transformer desgn. In case of a drect lghtnng flash to the MV lne, the surge arrester operaton or an nsulator spark-over dverts the surge current through the earthng system and can produce a resstve earth couplng between the MV and LV systems. An overvoltage s transferred to the LV system (as shown n the typcal case of Fgure 3-3). Dependng on the earthng mpedance values, ths earth couplng overvoltage can be much hgher than the capactve couplng through the transformer. In a TN system, f the neutral s also earthed at the customer nstallaton, smaller overvoltages wll occur. It should also be noted that ths knd of resstve couplng can be avoded by usng a separate earthng system for the LV part of the transformer. Fgure 3-3. case of overvoltage n the LV plant due to the couplng of the earthng wth the MV one. A typcal value of the overvoltage transmtted by capactve and electromagnetc couplng to the secondary of the MV/LV transformer sde s 2% of the MV phase-to-earth voltage between phase and neutral conductors and 8% between phase conductor and earth. In some partcular stuatons, ths MV/LV transfer rate can be hgher. Induced lghtnng surges on the MV system produce much less surge current (usually less than 1 ka) than drect flashes, and the overvoltages are practcally transferred to the LV system only by capactve couplng and do not exceed a few kv. In such cases, the overvoltage nduced drectly n the LV system s n general hgher than the one transferred from the MV sde

62 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3.2. Transmsson Lnes The Electro-Magnetc Transent Program, EMTP, s the envronment n whch the propagaton of transent dsturbances along the lnes of a dstrbuton network can be smulated and analyzed. The waveform of the voltages and currents at any pont of the power system are obtaned as analytcal soluton of the system of equatons correspondng to the combnaton of the equvalent model of the lnes and the boundary condtons mposed to the network. In EMTP envronment the three-phase lnes can be modelled by means of a constant-parameter lne model, also known as the Dommel lne model. In ths case the dstrbuted parameters of the lne R, L, and C are assumed to be constant. In partcular, they should be calculated for a frequency value representatve of the range of nterest. The model consders L and C to be dstrbuted ("deal lne"), whereas R to be lumped at three places (lne ends and lne mddle). One more dfference respect to the theoretcal lne model (reported n Fgure 3-4) conssts n neglectng the conductance G. The frequency dependence of the lne parameters s an mportant factor for the accurate smulaton of waveform and peak values. However, the cp-lne model s very robust and smple and provdes a good alternatve for a frst approxmaton analyss and for the modellng of secondary lnes. Fgure 3-4. Dstrbuted parameters lne model. The basc frequency doman equatons of the sngle phase dstrbuted parameter lne shown n Fgure 3-4,where the current and voltage values are reported n terms of phasors, are: dv ( x, t) di ( x, t) = R' I( x, t) L' (3.1) dx dt di ( x, t) dv ( x, t) = G' V ( x, t) C' (3.2) dx dt The prmed varables are gven n per lne length. When Laplace transformaton s used: dv ( x, s) = Z' I ( x, s) (3.3) dx

63 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS di( x, s) = Y ' V ( x, s) dx (3.4) where: Z' = R' + sl' Y' = G' + sc' Dfferentaton of equaton (3.3) and (3.4) results nto: 2 d V ( x, s) 2 = γ V ( x, s) 2 dx (3.5) 2 d I( x, s) 2 = γ I( x, s) 2 dx (3.6) wth: γ = ( R ' + sl')( G' + sc' ) = α + jβ (3.7) where α s the attenuaton constant and β s the phase constant. The general soluton of equatons (3.5) and (3.6) s gven by: V + γx γx ( x, s) = V e + V e (3.8) + γx γx [ V e V e c I( x, s) = 1 ] (3.9) Z Z R' + sl' = = ˆ θ G' + sc' c Z c Zc (3.10) Z c s the characterstc mpedance of the lne. Substtutng equaton (3.7) nto equaton (3.8) and assumng that V - and V + are phasors gves: V + jθ αx jβx jθ αx jβx ( x, s) = V e e e + V e e e (3.11) The tme-doman steady-state expresson of ths equaton s: + αx + +αx V ( x, t) = V e cos( ωt βx + θ ) + V e cos( ωt + βx + θ ) (3.12) Equaton (3.12) can be modfed and reused wth equaton (3.10) to convert equaton (3.9) nto tme-doman steady-state: + V αx + V +αx I( x, t) = e cos( ωt βx + θ θ Zc ) + e cos( ωt + βx + θ θ Zc ) Z Z c c (3.13) The term e -αx s the attenuaton of ampltudes of the waves. These expressons are the sums of forward travellng waves (+) and backward travellng waves ( ). A generc tme-doman representaton s wrtten as: + + V ( x, t) = V ( x vt) + V ( x + vt) (3.14) + I ( x, t) = I ( x vt) + I ( x + vt) (3.15)

64 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS wth wave velocty: v = ω. β The forward and backward travellng wave concept s nterpreted usng the llustraton n Fgure 3-5 for the waveform V + (x-vt). The travellng wave s frst shown at t = 0 where at x = a t has a value of V + (a). At any subsequent tme t x t has the same value at x = a+vt x (dstorton s neglected) as t formerly had at x=a. It means that the voltage dstrbuton has moved n the drecton of postve x. Fgure 3-5. The forward travellng wave at t = 0 and t = t x. A smlar explanaton s used for V whch s travellng n the negatve x drecton. In a lossless lne the voltage and current waveforms feature the same shape and ther ampltude depends on the characterstc mpedance of the lne; they travel along the lne wthout beng affected by dstorton. Moreover, current and voltage waves have the same polarty f they travel n the postve x drecton, whereas they have opposte sgn f they are travellng n the negatve x drecton. Reflected V 2, I 2 Incdent V 1, I 1 Refracted (transmtted) V 3, I 3 Z A Z B Fgure 3-6. Propagaton of voltage and current sgnals trough a dscontnuty pont. By referrng to Fgure 3-6 the reflecton and refracton coeffcents assocated to any dscontnuty secton of the electrc lne can be computed. Z A and Z B are the characterstc mpedances of the lne portons before and after the juncton, respectvely. The equatons gven by the contnuty condton n correspondence of the juncton n Fgure 3-6 are: V = 1 + V2 V3 I 1 + I2 = I3 But also the relatonshps between voltages and currents determned by the lne mpedance hold:

65 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS V1 I1 = ; Z A V2 I 2 = ; Z A I 3 = V 3 Z B Hence the reflected and refracted values are: V V Z B Z A = 1 Z + Z 2 V A B 3 2 Z B = V1 Z + Z A B = Γ α = Γ β (3.16) (3.17) Г α s the reflecton coeffcent and Г β the refracton one of the juncton between the two lne parts. In case of Z B < Z A, as when overhead lnes are connected to cables, the reflected voltage waveform s negatve whereas the reflected current waveform s postve (see Fgure 3-7). V 1 V 2 V 3 Z A Z B x I 1 I 2 I 3 Z A Z B x Fgure 3-7. Dscontnuty pont n the case of Z B < Z A. In case of Z B > Z A the reflecton coeffcent s postve, and the sgn of the reflected waveforms s opposte respect to the prevous case (see Fgure 3-8.). In the boundary condton of nfnte Z B,.e. when the lne s open at one end, Г α s equal to 1 and Г β s 2. By consderng the propagaton of the sgnals along the lne fed at the other end by an deal source (Z s = 0), as tme goes by a sequence of shots can be obtaned, as reported n Fgure

66 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS I 1 I 2 I 3 Z A Z B x V 1 V 2 V 3 Z A Z B x Fgure 3-8. Dscontnuty pont n the case of Z B > Z A. V I Source x Open end V 2V I Source x Open end

67 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS V I Source x Open end Fgure 3-9. Propagaton of current and voltage sgnals on a lossless open-ended lne. In practce, f T s the propagaton tme of the waves from one end of the lne to the other end, the value of the voltage at the open end and the value of the current fed by the source are the ones represented n functon of tme n Fgure 3-10 and Fgure v(t) 2V 0 T 2T 3T 4T 5T 6T t Fgure Voltage waveform regstered at the open end of the lne. s (t) I 0 T 2T 3T 4T 5T 6T t -I Fgure Current waveform generated by the source feedng the open ended lne. In the opposte condton,.e. n case of short crcut at one end of the lne, Z B s assumed zero, so the propagaton coeffcents relevant to that end are: Г α = -1, Г β =

68 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS V I Source x Short crcut V I 2I Source x Short crcut V 3I Source x Short crcut Fgure Propagaton of current and voltage sgnals on a lossless short-crcuted lne. In practce, f T s the tme spent by the waves for propagatng along the total lne length, the currents at the source end and at the short-crcut pont have the pattern represented n Fgure 3-13 and Fgure 3-14 respectvely

69 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS s (t) 5I 3I I 0 T 2T 3T 4T 5T 6T t Fgure Current waveform at the source sde of the short-crcuted lne. sc (t) 6I 4I 2I 0 T 2T 3T 4T 5T 6T t Fgure Current waveform at the short-crcut end of the lne. The deal operatng condton for a network s to have all the lnes matched, that means havng the same value of equvalent mpedance at both sdes of any dscontnuty pont. Such a condton s acheved by the presence of sutable adaptve resstance R m at the end of the lne, so that R m = Z A. In ths way Z A = Z B =R m and the lne has the same behavour of an equvalent nfntely long lne, n fact no reflected wave can occur and the transmtted wave s dentcal to the ncdent one (see Fgure 3-15). In the case of a real transmsson lne, losses have to be taken nto account; the man sources of energy loss are the non-zero resstance of conductors, the Corona effect and the leakage currents along nsulators. The effects of leakage and of longtudnal resstance are the exponental fall of the waveform front of voltage and current respectvely

70 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS V I Source x Matchng end V I Source x Matchng end Fgure Propagaton of current and voltage sgnals along a matched lne. In correspondence of the waveform front, anyway, the relatonshp between voltage and current stll holds: V I G t C = V 0 e (3.18) R t L = I 0 e (3.19) V = Z I (3.20) s The system of the three above equatons leads to the condton for a non-dstortng lossy lne: R G = (3.21) L C The energy dsspated on the characterstc mpedance Z B of a lne can be consdered as the energy transferred from lne A to lne B. the energy porton not transferred to the B- lne gves rse to a reflected wave. By the source-end perspectve, the lne behavour s the one of a pure resstance only f the lne s matched, but n general for the tme nterval 0 < t < 2T. R L G = = Z S I R = V G (3.22) C In a non-dstortng lne the energy can be thought as shared n equal parts between the electrc and magnetc feld, moreover the same porton of energy s dsspated by resstve losses and leakage

71 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS In case of mult-conductor lnes the propagaton of electrc sgnals depends also on the couplng between conductors. For the sake of smplcty let us consder two conductors and the ground, confgured as reported n Fgure M L L C 2 C 1 C 1 Fgure Two conductors lne - representaton of ts equvalent couplng parameters. dv di1 di 1 = L M 2 dx dt + (3.23) dt di dv d V V C C ) 1 ( 1 2 = (3.24) dx dt dt 1 The system of equatons above leads to the followng expresson: d V d V1 d V1 a b ) 1 2a + = 0 (3.25) dt dt dx dx ( 4 where: a = L ( C 1 + C ) MC 2 2 b = M ( C 1 + C ) LC 2 2 The solutons of equaton (3.25) have the followng expresson: V V 1 2 = f ( x + vt) 1 = f ( x + vt) 2 (3.26) where obvously v s a velocty. It can be shown that couple of travellng waves propagate along the lne at dfferent speeds: v v α 0 1 = ± a b 1 = ± a + b It can be sad that there are two dfferent propagaton modes:. conductor conductor;. conductor ground

72 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS In general, f the lne conssts n N conductors at a certan heght above the ground, N propagaton modes exst. In the case of two conductors,.e. N=2, f the ground s a perfect conductor (ρ=0) b s null, hence the propagaton speed s unque. If the ground s not a perfect conductor, on each lne conductor waves travellng at dfferent speeds are present; fnally, the resultng waveform s dstorted. Such resultng voltages and currents are just the phase sgnals Travellng Waves When a fault occurs along a transmsson lne, the voltage and current transents wll travel towards the lne termnals. These transents wll contnue to bounce back and forth between the fault pont and the two termnals of the faulted lne untl the post-fault steady state s reached. Evoluton of the termnal bus transents can be constructed by usng the Lattce Dagram method. Let us consder a sngle-phase lossless transmsson lne of length l, connected between buses A and B, wth characterstc mpedance Z c and travelng wave velocty ν. If a fault occurs at a dstance x from bus A, ths wll appear as an abrupt njecton at the fault pont. Ths njecton wll travel lke a surge along the lne n both drectons between the fault pont and the two termnal buses untl the post fault steady state s reached. A lattce dagram llustratng the reflecton and refracton of travelng waves ntated by the fault transents s shown n Fgure On the left sde of the fgure, a lne connectng buses A and B s vertcally drawn. A phase to ground fault s assumed to occur at pont F, x mles from bus A. The horzontal axs developng from pont F, corresponds to the tme reference. The arrows shown below the lattce dagram ndcate the arrval tmes of varous waves reflected from the fault as well as bus B. The travel tmes from the fault to bus A, and from the fault to bus B are desgnated by T 1 and T 2 respectvely. Gven the travelng wave velocty v, T 1 and T 2 wll be gven by: x T 1 = (3.27) v l x T2 = (3.28) v Constructon of the Lattce Dagram becomes computatonally dffcult f the attenuaton and dstorton of the sgnals are taken nto account as they travel along the lne. Ths s the case of lossy as well as non matched lne, as above explaned. In fact the reflecton and refracton coeffcents n the dscontnuty ponts depend on the characterstc mpedance of the lne and on the equvalent mpedance of source and load busses. On the other hand, tme-frequency analyss of the transent sgnals can be used to determne the propagaton tmes of these transents between the fault pont and the lne termnals

73 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS B T 2 F x t A T 1 T 1 +2T 2 3T 1 Fgure Lattce dagram for a fault at x mles from A. In three-phase transmsson lnes, f losses are taken nto account, there are three modes of propagaton, therefore for the analyss of the travelng wave effect, phase values must be converted nto modal values Modal Transformaton In three-phase systems the phase doman sgnals are frst decomposed nto ther modal components by means of the modal transformaton matrces: S = T (3.29) mode S phase where S mode and S phase are the modal and phase sgnals (voltages or currents) vectors respectvely. The lne currents transformaton matrx T s the matrx that makes dagonal the product Y ph Z ph where Y ph s the shunt admttance matrx n phase quanttes and Z ph s the seres mpedance matrx n phase quanttes. The resultng dagonalzng matrx, called T, determned by the egenanalyss routnes, s complex. To standardze the results, T s normalzed by usng the Eucldean Norm. The voltages transformaton matrx T v (whch dagonalzes the reverse product Z ph Y ph ) s not determned by the egenanalyss routnes, but calculated drectly from the relatonshp -t T v = T (where the superscrpt means nverse transposed). The matrces Y ph Z ph and Z ph Y ph are not equal but have the same egenvalues that, squared form the dagonal matrx [γ] 2. The elements γ of the matrx represent the propagaton constant of mode. They are complex numbers γ = α + jβ where α s the attenuaton constant and β s the phase constant of the mode. For a perfectly balanced lne, the modal transformaton matrces to relate modal and phase quanttes do not change wth frequency (constant transformaton matrces) and can be chosen to be real (e.g. generalzed Clarke). In partcular, Clarke's transformaton

74 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS s one of the commonly used modal transformatons for fully transposed lnes to decouple the phase doman sgnals nto ndependent modal components. One advantage of Clarke's transformaton s that, unlke the symmetrcal component transformatons, the correspondng transformaton matrx s real. Clarke's transformaton matrx, whch relates the modal vectors V 1,2,3 to the phase vectors V a,b,c s gven below: V V V = V 1 V 3 V a b c (3.30) If the studed lne s untransposed, then an egenvector based transformaton matrx, whch s frequency dependent, has to be used. Ths matrx should be computed at a frequency equal or close to the frequency of the ntal fault transents. For processng purpose of lne models n tme-doman real transformaton matrces T and T v can be used. To obtan approxmate T and T v matrces, the columns of T (complex) can be rotated to make the magnary parts of ts elements small and then retan only the real parts. In the case of the p-exact model, the fnal form of the model s expressed n terms of self and mutual phase quanttes, and there s no mpedment n usng exact complex transformaton matrces at each frequency at whch the model s produced. Ths model, however, s a one-frequency model, vald for steady-state solutons but not for transents smulatons

75 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3.3. Tme-frequency representaton of sgnals The analyss of non-perodc sgnals wth contnuous spectrum (as transent phenomena affectng power lnes are) s carred out by usng the wavelet transform, one of the most powerful tools for ths knd of study. The wavelet transform s based on the use of the tme-frequency representatons (TFR), whch allow to estmate how the sgnal spectrum s modfed vs. tme. Instead of transformng a tme-doman representaton nto a frequency-doman representaton, lke tradtonal operators (e.g. the Fourer transform), t uses a tme-frequency correlated descrpton. Ths way, t s easer to understand the tme nstant at whch an event occurs. Let us consder a generc sgnal x(t) expressed as a lnear combnaton of N sgnals, whose generc term s denoted by x k (t): N x( t) = a x ( t) (3.31) k = 1 k k Moreover, let us denote the tme-frequency representaton of x(t) by T x (t, f). The TFRs are classfed nto lnear and quadratc. As for the lnear ones, the followng relatonshp holds: x N T ( t, f ) = a T ( t, f ) (3.32) k = 1 k xk beng T xk (t,f) the tme-frequency representaton of x k (t). The quadratc TFRs are characterzed by a non-lnear combnaton of the terms T xk (t,f). For the sake of smplcty, but wthout loss of generalty, for N = 2 equaton (3.32) turns nto: T ( x1 2 x2 1 2 x1x2 1 2 x2x * * t, f ) = a T ( t, f ) + a T ( t, f ) + a a T ( t, f ) a a T ( t, ) (3.33) x 1 + f where the astersk refers to the complex conjugate and ( t, ), ( t, are cross T x f 1 x T ) 2 x f 2 x 1 terms due to the nterference between spectral components. The TFR of a sgnal, no matter whether t s lnear or not, lnks a one-dmensonal functon of tme (the sgnal x(t)) nto a b-dmensonal functon of tme and frequency T x (t, f). Hence, t can be represented by means of a 3-D surface, lke that depcted n Fgure Ths surface s the result of the applcaton of a lnear TFR to sgnal x(t) plotted n Fgure The Short Tme Fourer Transform (STFT) and wavelet transforms are typcal lnear TFRs, whereas the Wgner-Vlle Dstrbuton (WD) s a typcal quadratc TFR. By way of example, the man features of the STFT are brefly recalled n the followng. The STFT s a wndowed Fourer transform. To apply t, the observaton nterval s dvded nto a gven number of subntervals. Then, the STFT s computed, over each subnterval, accordng to the followng equaton:

76 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS X STFT - j2πfτ = x( τ) w( t τ) e d ( t, f ) τ (3.34) where w(t) s the wndowng functon, whch defnes the length of the subnterval and can be one of well-known (rectangular, Hannng, Hammng, ) analyss functons, and X STFT (t, f) s the result of the transform. x t [s] Fgure Aperodc sgnal. X(t, f) t [s] f [Hz] Fgure TFR of the sgnal x(t) shown n Fgure

77 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS The man drawback of the STFT s that the tme-frequency resoluton s constant and gven by the duraton of the wndow. However, t s essentally based on the well known Fourer transform and ths fact makes t easy and fast to mplement The wavelet transform The wavelet transform s a TFR representaton of a sgnal x(t). Ths knd of representaton s alternatve to the tradtonal Fourer-based transform and the use of one rather than the other s strctly dependent on the characterstcs of the sgnal to be analyzed. In practce, the wavelet transform becomes very useful when perodc sgnals wth short non-perodc components supermposed have to be studed. In other words, the wavelet transform allows good frequency resoluton at low frequences and good tme resoluton at hgh frequences. It means that t s possble to dscrmnate hgh frequency components very close to each other n tme and low frequency components very close to each other n frequency. Ths property s partcularly sutable for the study of the above knd of sgnals and, as t wll be clarfed soon, t s due to the nature of the functons whch form the bass for the wavelet transform. From an hstorcal pont of vew, the frst work where the man concept can be found was publshed at the begnnng of the XX century [2], but t s n the 80s that the wavelet transform starts to become very popular, see e.g. [3-7], and appled n many felds, such as power qualty (manly for the characterzaton of electromagnetc transents), dagnostcs, mage processng, data compresson, bomedcne, geophyscs (by way of example, see [8-13]). The Contnuous Wavelet Transform (CWT) X(a,b) of a functon x(t) s defned as follows: t b X ( a, b) = x( t) w( )dt = x( t) wa,b ( t) dt (3.35) a a where w a,b (t) s the wavelet derved from a functon w(t), whch s referred to as mother wavelet, by changng the parameters a (scale) and b (tme). The mother wavelet w(t) s usually referred to as w 1,0 (t). The factor 1/rad( a ) ensures that w a,b = w = 1. Fgure 3-20 shows the plot of the mother wavelet of Morlet. The mother-wavelet frequency spectrum s smlar to the one of a band-pass flter. Parameter a both vares and shfts the frequency bandwdth (narrow band at low frequency, wde band at hgh frequency). In the tme doman, t corresponds to a dlataton or a contracton of w(t). Parameter b smply operates a tme translaton of the mother wavelet. The so-obtaned functons (wavelets) are the bass for the transform. In ths way, x(t) s analyzed wth dfferent resoluton n frequency and tme, and s decomposed nto sub-bands havng bandwdth scaled by a

78 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS A generc functon must meet two smple condtons n order to be defned as mother wavelet:. beng well localzed n tme and. have zero mean. However, further propertes such as orthogonalty and borthogonalty are, n practce, requred, as explaned n the followng. Fnally, t s worthwhle emphaszng that the wavelet transform of a sgnal x(t) s the nner product between the sgnal tself and the wavelet w a,b (t): X(a,b) = x(t) w a,b (t) It means that the coeffcents X(a,b) are ndces of the smlarty between x(t) and w a,b (t). Ampltude Ampltude Samples a) b) Harmonc Order Fgure Mother wavelet of Morlet (a) and the relevant magntude spectrum (b). Hence, a functon w(t) whch s good for the analyss of a gven sgnal, could not be as much useful for another one. The choce of the more fttng mother wavelet s stll an ssue when applcaton of the analyss s concerned; studes are at present n progress amed at developng sutable algorthm whch automatcally select the mother wavelet n functon of the sgnal to be analyzed. To better clarfy ths pont, let us consder Fgure 3-20 and Fgure Fgure 3-20b depcts the magntude spectrum of the wavelet of Morlet of Fg. 4.5 and Fgure 3-20a, whch s, as above mentoned, qute smlar to the one of a band-pass flter. By decreasng the value of a, a contracton n the tme doman occurs, as shown n Fgure 3-21a. Fgure 3-21b shows the effects n the frequency doman: the spectrum s shfted to hgher frequences and the bandwdth s ncreased. By decreasng agan the value of a, the functon s some more contracted and the spectrum s some more shfted and ncreased. It s clear that, f a s ncreased, the behavor of the functon s exactly reversed: dlataton n tme vs. contracton and down-shftng n frequency

79 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Ampltude Ampltude Samples a) b) Harmonc Order Ampltude Ampltude c) Samples d) Harmonc Order Fgure Wavelet obtaned by contractng the mother wavelet of Morlet (a, c) and the relevant magntude spectrum (b, d) The dscrete-tme wavelet transform The CWT defned by (3.35) cannot be used for practcal applcatons because t cannot be mplemented on DSP systems. The dscrete-tme wavelet transform (DTWT) s the dgtally usable verson of the CWT. It s defned by the followng expresson: = m 1 na0 X m, x[] n w[ ] m n m a a0 0 (3.36) Where x[n] and w[ ] refer to the dgtzed nput sgnal and the dscrete counterpart of the mother wavelet, respectvely. Parameters a, b have been expressed by the followng functons: a = a m 0, b = na = na m 0, where a 0 and m are nteger and m refers to the generc m-th frequency sub-band of the sgnal. Typcally, t s chosen a = 2, thus allowng to analyze the sgnal n a hgh number of sub-band. In order to present how the DTWT s usually mplemented, let us dscuss the followng example taken from [14], n whch DTWT s based on the Haar bass [2]. Let us consder

80 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS the followng elements of x[n]: x[0], x[1], x[2], x[4]. The transform replaces these four numbers wth other four numbers y[0], y[1], y[2], y[4] that are, due to the lnearty, a lnear combnaton of the formers. The followng sums and dfferences are made: y[0] = x[0] + x[1] y[2] = x[2] + x[3] y[1] = x[0] - x[1] y[3] = x[2] - x[3] If the values x[0] = 1.2, x[1] = 1.0, x[2] = 1.0, x[4] = -1.2 are consdered, the followng results hold: y[0] = 2.2, y[1] = 0.2, y[2] = -2.2, y[3] = 0.2. It should be noted that the value of y[1] and y[3] can be consdered neglgble wth respect to y[0] and y[2]. Hence, one can conclude that, n ths new representaton, only two terms are suffcent to represent ths sgnal and compress t by cancellng y[1] and y[3]. Obvously, n such a case we are not able to perform a perfect reconstructon of x. Moreover, t should be underlned that y[0] and y[2] are obtaned by summng consecutve nput samples and, hence, they are a knd of movng average of the nput sgnal. It means also that they are the result of a low-pass flter. As for y[1] and y[3], they are gauged by the dfference of consecutve nput samples: hence, they are a knd of movng dfference of x. Ths s the typcal result of hgh-pass flter. In partcular, ths flter s the mrror of the above low-pass flter. Summarzng, the applcaton of the transform has allowed to dvde the hghest frequency components, whch are referred to as detals, from the lowest ones, whch are referred to as approxmatons. Fgure 3-22 depcts ths stuaton. approxmaton y[0], y[2] y[1], y[3] detals f max 2 f max f Fgure Detals and approxmaton n the frequency doman. a samplng frequency equal to f s was assumed and hence f max s the Nyqust frequency f s /2. If the transform s then appled to these approxmatons, other detals and approxmatons wll be gauged: z[0] = y[0] + y[2] = 0 z[2] = y[0] - y[2] = 4.4 The term z[0] s the approxmaton (lower component) whereas z[2] s the detal (hgher component). Fgure 3-23 depcts ths result

81 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS approxmaton z[0] z[2] detal f max 4 f max 2 f Fgure Detals and approxmaton n the frequency doman after the teraton of the procedure. Moreover t should be noted that the component z[2] does most of the work of the orgnal four. Hence, a strong compresson of the data can be obtaned. The descrbed procedure turns nto a mult-resoluton analyss. The meanng of ths term becomes clear f Fgure 3-24, whch s gauged by combnng Fgure 3-22 and Fgure 3-23, s observed. approxmaton z[2] detal z[0] y[1], y[3] detals f max 4 f max 2 f max f Fgure Sub-bands obtaned by the teraton of the low-pass and hgh-pass flters. As a result, the nput sgnal has been analyzed by means of band-pass flters havng bandwdth scaled by power of 2. By keepng n mnd, as mentoned n the prevous Secton, that wavelets are essentally sutable band-pass flters, one can conclude that the DTWT can be mplemented by followng a procedure lke ths. At the end of ths example, two consderatons must be ponted out: t s possble to mplement the DTWT by means of a cascade of proper low-pass and hgh-pass flters (flter bank); the wavelet transform can be used also to compress data, snce t represents the sgnal wth many neglgble terms Flter banks A flter s a lnear tme-nvarant operator whch acts on an nput vector x and produces an output vector y gven by the convoluton of x wth a gven vector h [14]. Ths vector contans flter coeffcents h[0], h[1],. It holds y [ n] = h[ k] x[ n k] (3.37) k

82 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS h[0], h[1], are the elements of the mpulse response, obtaned by applyng a unty ampltude sample. In the frequency doman can be rewrtten as: Y ( ω) = H ( ω) X ( ω) (3.38) Where X H + - jπω ( ω) = x[ n] e (3.39) n= + - jπω ( ω) = h[ n] e (3.40) n= A flter bank s a set of flters, whch can be dvded n analyss bank and synthess bank, and has often two flters, low-pass and hgh-pass. The analyss banks splt the nput sgnal nto frequency bands whereas the synthess one allows the sgnal to be recombned. Fgure 3-25 shows a typcal two-channel flter bank. H 0 and H 1 are a lowpass and a hgh-pass flter, respectvely, whereas F 0 and F 1 are the correspondng flters of the synthess bank. x[n] H 0 2 v 0 [n] 2 F 0 H 1 2 v 1 [n] 2 F 1 x s [n] Analyss Synthess Fgure Analyss and synthess block dagrams of a two-channel flter. As for symbols 2 and 2, they ndcate a decmator and an expander by 2, respectvely. The length of the output of the analyss bank s doubled, snce both v 0 and v 1 have the same length as x. Ths nformaton s redundant and hence t s possble to down-sample the outputs by removng, for example, the odd-numbered elements. Ths s a lnear operaton but t s not tme-nvarant. The flterng and decmaton operatons can be done wth one matrx only where the down-samplng s performed by removng the oddnumbered rows of the flter matrx and hence, n the resultant matrx, whch becomes rectangular, each row can be obtaned from the prevous one wth a double-shftng. As for the expander 2, t performs an up-samplng by nsertng zeros n odd-numbered elements. To do ths wth a matrx, zero rows have to be nserted. In other words, each column can be obtaned from the prevous one wth a double-shftng. When x s [n] s equal to x[n], wth at the most a delay l, the flter bank performs a perfect reconstructon. The followng two condtons have to be met for a perfect reconstructon: F 0 (z) H 0 (z) + F 1 (z) H 1 (z) = 2z -1 (3.41) F 0 (z) H 0 (-z) + F 1 (z) H 1 (-z) = 0 (3.42)

83 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS The frst condton states that there s no dstorton, whereas the second condton determnes alas cancellaton. Moreover, the latter one provdes smple relatonshps between the flter coeffcents (alternatng sgns): F 0 (z) = H 1 (-z) and F 1 (z) = -H 0 (-z) (3.43) A flter bank whch performs perfect reconstructon s borthogonal. It means that the flters of the synthess bank are the nverse of flters of the analyss bank. Moreover, when the synthess bank s the transpose of the analyss one, the flter bank s orthogonal. In such a case, a further condton on flter coeffcents arses: h 1 [k] = (-1) k h 0 [N-k] (3.44) where N s an odd number denotng the number of taps. Hence, under orthogonal condtons, fxng the coeffcents of the low-pass flter allows defnng all flter banks by means of (3.43) and (3.44) Wavelets and flter banks As a concluson of the above dscusson, t should be clear that the DTWT can be mplemented by means of the analyss flter bank, such as the one shown n Fgure called wavelet decomposton tree. In ths pcture, h and g are the mpulse response of a sutable hgh-pass and of ts dual low-pass flters, respectvely. The number γ and the value of the coeffcents of each flter are typcal of a gven mother wavelet. These flters have three man characterstcs: I. they are FIR flters; II. the sum of ther coeffcents s 1; III. ther norm s 1. The generc m-th stage of the flter bank, whch performs a parallel flterng and a downsamplng of the flters output, produces two output vectors y m and β m, whch are referred to as detal coeffcents and approxmaton coeffcents, respectvely. Each vector β feeds the followng stage. It concdes wth the nput sequence x[n] when the frst stage s consdered. The detal vectors y 0, y 1,..., y m,... represent the sgnal n the sub-bands 0, 1,, m,, respectvely. Both vectors y m and β m relevant to the m-th sub-band can be expressed, as suggested n [14], by means of the followng matrx form:

84 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS.. βm[ j]. y m[ j]. β m[ j + 1] =. y + m[ j 1] g[0] h[0] g[1] h[1] g[2] h[2] g[0] h[0]..... g[1] h[1]... g[ γ 1] h[ γ 1] g[2] h[2] g[ γ 1] h[ γ 1] βm-1[ j]. β m-1[ j + 1]. β -1[ j + 2] (3.45) m. β [ + 3] m-1 j.... When the mother wavelet gves rse to orthogonal bass, also the matrces (and the flter bank) are orthogonal. The followng relatonshp between taps holds: g[n] = h[γ-n-1] n = 0, 1,, γ-1 (3.46) It must be noted that an deal flter bank cuts the frequency band n half and hence there s no overlap between the sub-bands. But n practcal applcatons actual flters wth a fnte number of coeffcents are used and hence overlappng of sub-band arses. Ths causes a sharng of nformaton between adjacent sub-bands, makng hence less effectve the flterng propertes of such a transform. x[n] h 2 y 0 g 2 β 0 h g 2 2 β 1 y 1... stage h 2 y L g 2 β L Fgure Flter bank for the DTWT mplementaton Thanks to ts nterestng propertes of mult-resoluton analyss, whch allows to get good frequency resoluton at low frequences and good tme resoluton at hgh frequences, wavelets are wdely appled n power qualty, specally to the study of voltage transents. Hence, several papers relevant to ths applcaton can be found n the lterature. Although t started to be used n mage processng up to the last few years, t s only n 1993 that the wavelet transform s assocated to power qualty and n 1994 the frst results are publshed where some voltage dsturbances are analyzed by means of the CWT. It must be noted that dfferent knds of Daubeches mother-wavelet (wth 4, 6, 8 and 10 coeffcents) allows the most part of voltage transents to be correctly analyzed. In

85 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS partcular, db4 (Daubeches wth 4 coeffcents) and db6 seem to be useful for fast transents, whereas db8 and db10 become more attractve n the presence of slow transents

86 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3.4. Fault locaton methods for dstrbuton networks Fault locaton s mportant for safe operatng condtons of power systems. In transmsson power systems the development of technques for estmatng the locaton of a fault along the lnes has been consdered an essental requrement for power companes. Despte somehow useful, lttle work has been done n the development of fault locaton technques for dstrbuton networks, untl recent past; n ths sense fault locators have been consdered not essental on dstrbuton systems. Nowadays, ths ssue has become one of the man subjects n dstrbuton companes, as a consequence of the trend n regulaton and prcng towards effcency and power qualty mprovements. Hence, the mportance of accurate fault locaton n dstrbuton networks s ncreased. Wlls [15] presented a resume of some of the classcal technques that have been used through the years to locate cable faults n dstrbuton systems. A general but useful reference s a report of the CIRED workng group [16], where the problem of fault locaton s presented as one of the man ssues n a fault management system. Tang, et al. [17] revew the fault ndcator applcatons both n transmsson and dstrbuton systems; prncples, merts and demerts of fault locaton technques are dscussed. In electrcal dstrbuton networks one of the man ssues s the fault locaton for repar and nspecton purposes, because fault management s essental to reduce outage tmes. It has not so much to do wth protectve relayng, snce the protecton scheme s an on-lne applcaton and the speed of operaton s the man ssue, whereas the fault locaton technques are usually off-lne applcatons n whch accuracy aspects are the man concern. Fault locaton n dstrbuton systems dffers consderably from the approaches appled for transmsson systems. In fact the sgnfcant dfferences n networks structure, dmensons and groundng prncples must be consdered when a fault locaton method s used n dstrbutons systems. It must be ponted out that many of these characterstcs have not been totally consdered n the bblography for example most of the fault locaton approaches for dstrbuton systems are developed for soldly grounded systems, a lttle percentage of publcatons consder and apply successfully the method to a feeder wth multlaterals, and, last but not least, very rare are the technques that can be used n case of networks consstng n both overhead lnes and cables. Most of the fault locaton methods were developed for transmsson systems and are not sutable for radal dstrbuton networks. Tradtonally, the short crcut faults n power dstrbuton lnes were located by tral and error method,.e. by dvdng the lne nto sectons and tryng to close the energzng crcut breaker: The approach takes a long tme and stresses the equpment even more than the fault tself. A fault can thus be located n several teratons, ts man drawback s the tme requred to localze the

87 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS faulted secton and the fault. Crew productvty can be mproved through the use of faulted crcut ndcators,.e. sngle or multphase devces desgned to sense the fault current flowng through the power conductor(s) at the pont where ther sensors are nstalled and provde an ndcaton of the event [18]. Nowadays consderable research efforts nto development of new fault locaton technques for dstrbuton systems have been spent; the technques can be classfed under two man categores, reported n the dagram of Fgure Fault Detecton and Locaton Methods for Electrc Power Systems Sgnal analyss of current and voltage acqured n strategc ponts Procedure based on the knowledge of the network and typcal patterns Hgh-frequency content travellng waves theory Fundamental frequency components: phasors artfcal ntellgence: expert systems, neural networks genetc algorthms Sngle ended measurements Double ended measurements Fgure Dfferent approaches present n scentfc lterature for fault locaton n dstrbuton systems In recent years sgnal analyss approaches ncreased a lot. In partcular, wth the rapd developments of mcroprocessor technology, research actvtes focused on the hgh frequency components of voltage and current sgnals. Fault locaton technques usng sgnal analyss are based on measurements of current and voltage. These technques can be dvded nto two categores: methods analyzng the fundamental frequency components and methods analyzng the hgh frequency components. The latter methods can be further classfed nto two sub categores, accordng to the avalable measurements: so-called sngle ended or double ended methods. Sngle ended technques use only one termnal measurements, and are based most of the tmes on the nformaton provded by dgtal fault recorders nstalled at the head of the feeder. the advantage showed by ths approach s that communcatons are not needed. The double ended technques gve more relable and accurate results, but communcaton channels have to be ncluded n the measurement system. The fact that a feeder has many branches makes the locaton of each fault much more dffcult, n fact the voltage and current sgnal analyss yelds to obtan more than one fault pont canddate wthn the network

88 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Sgnal analyss of fundamental frequency components The fault locaton technques based on fundamental frequency use the power frequency phasors to estmate the fault dstance n dstrbuton systems. Tll now dfferent algorthms for sngle-ended fault locaton methods have been proposed, where a trend towards mpedance measurement technques s shown. The value of the apparent mpedance must be compensated at each fault occurrence of the contrbuton correspondng to the fault resstance, hence such methods have lmted applcatons because they cannot provde the requred accuracy n the locaton when system or fault condtons change. In 1993, A. Grgs et al. [19] presented a sngle-ended fault locaton technque for rural dstrbutons feeders based on phasors obtaned wth a recursve optmal estmaton algorthm. The dstance from the head of the feeder to the fault s then estmated on the bass of the apparent mpedance computed by usng updated values of voltages and currents. The approach [19] has then been consdered as a reference pont for the development of some later fault locaton methods, for example [20]. In ths paper the authors ncrease accuracy of the fault locaton, provded there s no remote n fed. Upon the detecton of a dsturbance, phasor quanttes of voltage and current are obtaned by an estmaton program; the varaton n the magntudes of the phasors s used as the crtera to classfy the type of fault and dentfy the faulted phases. In partcular, the relatve change n the magntude over the reference value of each phase current s computed and compared wth a threshold value n order to detect the faulted phases. After the classfcaton of the type of fault, a par of voltage and current phasors s chosen to compute the relevant apparent power. The fault boundary condtons and the sequence network parameters are used to calculate the sequence components of that voltage and current, so that the equvalent apparent mpedance s found. A compensatng current s consdered to be fed nto the fault pont to elmnate the fault resstance contrbuton to the apparent mpedance. The obtaned value of apparent mpedance s used to estmate the dstance from the fault locaton. In ths approach, to overcome the lack of nformaton about the fault resstance, t was assumed that the current contrbuton due to the fault s the change n the current magntude from pre- to post-fault. In 1997 R.K. Aggarwal, et al. [21] presented a sngle-ended fault locaton technque for overhead dstrbuton systems, based on the concept of supermposed components of voltages and currents, defned as the dfference between the total post-fault and pre-fault quanttes. By usng a specfed model of the dstrbuton system, voltage and current values of the supermposed components are computed so that a supermposed components crcut can be consdered. In ths equvalent crcut the poston of the fault s systematcally vared along a lne n order to deduce the pont gvng the mnmum value of supermposed current n the healthy phase(s). In ths way the fault locator can be

89 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS hghly nsenstve to varatons n local and remote source mpedances, to the fault resstance and to the presence along the feeder of taps wth varable loads. In 2003, P.M. Van Orsouwf, et al. [22] reported the experence of the Dutch dstrbuton company about a reactance based methodology for fault locaton. The system revealed to be able to locate two- and three-phase faults wthn the requred accuracy of 100 meters and sngle phase faults wthn 1000 meters. 5 mnutes after the fault occurrence the dspatch center could communcate to the emergency crew where the fault should be checked, hence the company could save around one hour to look manually for the fault Sgnal analyss of hgh frequency components The use of non-power frequency components for power system analyss has been contemplated for a long tme, especally n power system protecton, but only n the last years has been sgnfcantly developed. Recently, travellng waves methods, almost used n the past for protecton purposes n transmsson lnes, are back as an alternatve approach for fault locaton. The theory behnd these methods s based on the correlaton between the forward and backward travellng waves along the lne. These methods montor the correlaton coeffcent between hgh frequency waveforms, whch ncreases sgnfcantly n case of fault. The transent generated by the occurrence of a fault wll get to the measurement pont and then wll be reflected back towards the fault pont; the same happens at the fault poston, fnally the transent wll arrve at the relay termnal as a hghly correlated sgnal after a tme delay equal to twce the travellng tme of the transents from the relay to the fault locaton. Ths tme nterval can then be used to estmate the dstance of the fault pont from the relay termnal. However, for dstrbuton systems, the problem becomes more complex because the topology n radal grd nvolves many reflectons. Unfortunately, t s well known that the travellng waves based method does not perform well for certan type of faults and/or system condtons. For example, faults very close to the measurement pont as well as faults occurrng close to a zero crossng of the steady-state voltage waveform. One of the frst reports on the use of hgh frequency components to determne fault locaton n dstrbuton systems was proposed n 1992 by A.T. Johns [23]. The technque s qute robust respect to varatons of the type of fault, of the value of fault resstance, of the source short-crcut level and to the pont-on-wave of fault occurrence. The man drawback s that to obtan such performance the locators need to be nstalled at strategc and convenent regular ntervals along the overhead dstrbuton network. In 1998, Z.Q.Bo presented the applcaton of transent based fault locaton technques to the dstrbuton systems by utlzng GPS [24]. It reles on detectng hgh frequency sgnals generated by faults and overcomes the longstandng ssue whch restrct the use of travellng wave technques: the dentfcaton of multple reflectons from the busbars and

90 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS the fault pont. The fault locaton s performed by comparng the dfferences between nstances n tme for all locators n the system. In [25], despte the complexty of dstrbuton systems, the authors show that t may be possble to obtan accurate fault locaton even by applyng standard and well known technques developed for EHV networks. The fundamental dea presented n the paper s that the transent dsturbance travels on every transmsson lne and gets to the substaton busbars, generatng agan transmtted and reflected transents. Whle on the faulted lne there wll be both ncdent and reflected pulses, on all the other safe transmsson lnes only the transmtted transent should be found. Hence, by measurng the currents affected by the fault transent, the voltage on the lnes s estmated consderng only one propagatng wave. Fnally, to locate the fault, arrval tme of each ncdent wave at the measurement pont s dentfed by means of the cross correlaton between reflected and ncdent waves. The man drawback of the prevous proposals s that they are based on a double-ended method and due to nvestment and nstallaton costs these schemes are costly respect to the effects of faults n dstrbuton systems. A more attractve approach s a fault locaton method where current or voltage sgnals are measured at just one end of the lne (sngle-ended), usually at the substaton. In ths case fault locaton reles on the analyss of the sgnals to detect the reflectons between the measurng pont and the fault. One of the man problems s the presence of multple possble locaton of the fault for a gven recordng at the measurement pont. One of the frst attempts n ths lne was proposed n 1999 by F. Magnago, et al. [26]: the analyss s performed usng the transent sgnals recorded at the substaton durng the fault. The method dentfes the faulted lateral, frstly estmatng the dstance from fault pont accordng to the travellng wave nformaton provded by the hgh frequency components of the event, secondly by usng the specal propertes of wavelet transform coeffcents to fnd the rght poston wthn all the faults that can occur along dfferent laterals of the same man feeder, equally dstant from the man substaton. Then, the fault locaton along the dentfed lateral s computed on the bass of a smplfed network model and on the steady-state phasors as n the methodology [19]. Unlke the correlaton-based methods, where the forward and backward travellng wave components are computed and used for the cross-correlaton, n the wavelet based approach the entre sgnal acqured at the relay poston n the network s drectly analyzed. Transents along multphase transmsson lnes can be analyzed by usng the modal transformatons, decouplng the travelng wave equatons nto ndependent equatons for each mode. Voltage and current sgnals at the lne termnals can be transformed nto

91 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS modal components, and each mode can then be analyzed separately. Frst mode s usually referred to as ground mode, second mode s also known as aeral mode and can be found for any knd of fault. The sngle-ended recordng method turns nto two dfferent technques dependng on the fact that the fault s grounded or ungrounded. If the short crcut s a lne-to-lne or a threephase one the fault s ungrounded, hence not sgnfcant reflectons are caused by the remote-end bus durng the transent. In such cases, the dstance from fault to recorder can be easly computed on the bass of the tme delay between consecutve peaks by usng the wavelet transform coeffcents at the frst decomposton level. When the short crcut nvolves a connecton to ground, the second approach s followed: sgnals can be affected by a sgnfcant reflectons contrbuton due to the remote-end bus superposed to the ones comng from the fault locaton. Therefore, other levels of coeffcents of the wavelet decomposton are analysed n order to confrm that the fault s grounded, then the frst decomposton level s consdered to understand n whch half part of the lne the fault can be found, and fnally the rght expresson for the dstance of the fault s chosen and appled. In 1999, Z. Q. Bo et al. [27] takes nto account the man ssues n the state of the art relevant to fault locaton methods: on the one hand mpedance based methods, beng power-frequency based measurement methods, suffer from lmtatons due to the system condtons: fault path resstance, lne loadng, source parameters, and so on. On the other hand, travellng wave based methods exhbt shortcomngs: the fault occurrence does not generate many travellng waves components when t corresponds to a phase angle near zero of the 50 Hz voltage component; faults close-up to the measurement staton gve rse to ncdent and reflected waves hard to be separately detected. Do not forget the lmted bandwdth of the current and voltage transducers feedng the measurement system. In order to overcome most of the above ssues, the authors present a technque relyng on the detecton of the hgh frequency sgnals generated by the fault arc. These components don t vary wth the pont of the voltage waveform at whch the fault occurs; moreover, multple reflectons from busbars and the fault pont are not needed to be detected. The prncple of the fault locator s the dentfcaton of the arrval sequence of the transent hgh-frequency voltage sgnals (1 10 MHz) at the busbar where the locator s nstalled. The busbar mpedance s domnated by capactance effect, thus the reflected wave comng from the remote busbar should have opposte polarty respect to the one comng from the fault pont: a dstncton between reflecton ponts s obtaned. In ths case the accuracy of the locaton s proportonal to the samplng rate of the locator devce

92 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Peak magntudes of the sgnals are gnored, n favour of ther polarty and arrval tme detecton. In 2002 P.F. Gale, et al. [28] show that t s possble to locate a fault on a dstrbuton feeder consstng of underground cables, overhead lnes and tapped load, usng measurement at a sngle locaton. The magntude of the reflected and transmtted components depends on the varaton of the equvalent mpedance n dscontnuty ponts of the power system. Identfcaton of the desred sgnal s of crucal mportance for the correct operaton of the fault locaton method. A probable locaton of the fault s estmated by comparng the relatve dstance of each peak n the hgh frequency current sgnals to the known reflecton ponts n the dstrbuton feeder. The estmated fault locaton s then used wthn a transent power system smulator that models the actual network. The resultng smulated current waveforms are then cross-correlated aganst the sgnal captured on the real network: f the poston of the fault s correctly estmated, hgh frequency components n the smulated waveform wll be smlar to that of the measured waveform, and thus the obtaned cross-correlaton value wll be hgh postve. It must be notced that n ths approach the cross correlaton s carred out not on the bass of the Bewley lattce dagram, very dffcult to follow when the fault resstance s hgh and n general because of the multple reflectons travellng from remote ends of the laterals to the montorng pont. The tme trees technque has been used for waveform predcton, whch gves a good vsual descrpton of the way waveforms are generated at each end of the network, by consderng the travellng path from the fault poston towards the dscontnuty ponts along the network. Nucc et al. proposed n [29] a procedure based on the contnuous wavelet transform for the analyss of voltage transents due to lne faults. The analyss, carred out by applyng the method for fault locaton n dstrbuton networks, shows that there s correlaton between some characterstc frequences of the transformed sgnal and specfc paths along the network covered by the travellng waves generated by the fault. The correlaton provdes useful nformaton for the locaton of the faulted secton by usng the recorded transents at one or more busses. In ths paper the contnuous wavelet transform s used nstead of the dscrete wavelet one, so that a more detaled analyss of the spectrum energy of the voltage transent can be performed. The contnuous wavelet transform can operate at any scale, and s also contnuous n terms of shftng. The dstnctve frequences of propagaton and reflectons of the fault transent are detected. By knowng the network topology and the propagaton speed of the sgnals, the locaton of the fault can be nferred

93 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Artfcal neural networks technques Artfcal ntellgence technques, such as expert system, neural network, fuzzy logc and genetc algorthm, have been employed to fault locaton n dstrbutons systems. The use of artfcal neural network does not requre the formulaton of the soluton algorthm, because t should be able to employ mplct dependences on varous parameters n the tranng data. These technques are based on the assumpton that durng the fault the voltage and current sgnals measured by the network unts contan nformaton about the locaton of the fault. The unquely defned fault pattern can be recognzed by the ad hoc desgned neural network. Intal attempts (1991) at the applcatons of expert system for fault locaton n dstrbuton network can be found n H. Yuan-Yh, et al. [30], where an expert system s proposed to help the dspatchers to locate the faults n dstrbuton systems, the effectveness of the desgned expert system s demonstrated on a dstrbuton system wthn the servce area of the Tape Cty Dstrct Offce of Tawan Power Company. In 1996, Jarventausta, et al. [31] proposed an expert system that combnes nformaton obtaned from the network database and heurstc knowledge of operators amng at obtanng potental fault locatons. The electrcal dstance between the feedng connecton and the fault pont s determned by comparng the measured short crcut current and the type of fault wth the calculated fault currents of each lne secton. The nformaton obtaned from the remote control system s affected by varous uncertanty contrbutons and may be wrong, partally msleadng or not suffcent to the purpose. A more nterestng method than the conventonal solutons for fault locaton n dstrbuton system s the use of artfcal neural network (ANN), snce t does not requre the explct formulaton of the soluton algorthm, but s able to mplctly employ varous dependences n the tranng data. The technque s based on the assumpton that durng the fault, before t s cleared, the voltages and currents measured by the network unts contan nformaton about the locaton of the fault n the system. These measured quanttes unquely defne a fault pattern (or sgnature) that can be recognzed by a specally desgned neural network. The challenge n ANN fault locaton technque s to be able to process the data for a gven pattern and relably extract the nformaton about fault locaton. A comparson of ANN technques for fault locaton was carred out n 1996 by J.A. Momoh [32]. The authors conclude that hgher success rates are achevable usng the back propagaton method and the clusterng method when compared wth the counter propagaton method

94 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Probablstc methods are other approaches for fault locaton, for example, n 2000, S. Hannnen, et al. [33], proposed a probablstc method based on the change of the neutralvoltage and zero-sequence currents. Concludng the survey, t can be sad that no technque features all the desrable features that lead to the accurate fault locaton n dstrbuton systems. On the one hand tradtonal fault locaton technques n whch fundamental voltages and currents at lne termnals are consdered, are dffcult to be appled n dstrbuton lnes wth tapped loads. On the other hand, Travellng waves based fault locators are sgnfcantly more relable f a doubleended method s used, when sngle-ended fault locaton should be cheaper and hence preferred n case of dstrbuton systems. Nowadays research nterests and actvty are actually tryng to develop a novel fault locaton method, based on a central fault locator nstalled n the substaton wth only currents and voltages measurements avalable. To do ths the sgnal-analyss based method and the knowledge-based method can be convenently combned n order to develop hybrd systems that could overcome lmtatons of the sngle solutons. Hence, there s a techncal challenge n research and development for the successful desgn and mplementaton of a practcal fault locaton method for dstrbuton systems

95 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3.5. Dstrbuted measurement system detectng transent dsturbances and method to locate the transent source In the contrbuton [34] the electrc and electronc measurement group of Bologna presented a dstrbuted measurement system for power qualty montorng. In partcular, the use of such a system was dscussed n connecton wth a lghtnng locaton system amng at nvestgatng on the correlaton between mpulsve transents and ther nducton by strokes. On the bass of the same dstrbuted measurement system a new functon s proposed,.e. the detecton of transent events affectng the lnes sgnals and the locaton of the source of dsturbance wthn the montored dstrbuton network. The measurement system conssts n a master-slave archtecture, where all the slave unts have the same tme reference and can communcate nformaton to the master unt. The method for the real-tme locaton of the fault or any other transent source reles on the measurement of the startng nstant of the dsturbance at gven ponts of the network, hence t does not nvolve any sgnal analyss. Input voltage u VVT u VVT EDB u out GPS-based devce PC Startng nstant Transent parameters Trgger n DAQ Fgure Block dagram of a slave unt channel. Fgure 3-28 s the block dagram of one of the three measurement channels of a remote unt (slave). The hardware ncluded n the staton wll be descrbed and metrologcally characterzed n chapter 5. As a transent dstorton affects the nput voltage, the system acqures both the startng nstant and the waveform of each transent. The former nformaton s then sent to the master unt whch locates the source of the transent by processng the data receved from all the slaves. The dashed box n Fgure 3-28 contans the blocks performng the measurement of the transent startng nstant. Only by way of example, n ths case the three phase-sgnals chosen for the network montorng are the lne-to-ground voltages. In each channel the voltage u(t) at the montored pont s condtoned by a Voltage-to- Voltage Transducer (VVT), whose output u VVT feeds an Event Detecton Block (EDB). The EDB output u out s a logc sgnal that, as a transent occurs, trggers both a Data AcQuston board (DAQ) and a GPS-based devce, whch provdes the relevant tme

96 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS stamp. The tme stamp s determned wth an uncertanty that propagates through the algorthm whch the proposed measurement procedure reles on, and affects the estmate of the transent source locaton. In chapter 5 also the combned uncertanty affectng the locaton s estmated. In the same way the measurement system can also locate the source of transent currents. In such a case, n Fgure 3-28 the output of a sutable current-to-voltage transducer n place of u VVT has to be consdered Procedure to locate the source of a transent The method requres that the characterstcs of the consdered network (topology, geometry of the lnes, type of conductor and space between them) are known. Such a requrement s not a problem n practcal mplementatons. Snce the method reles on the travellng waves theory, reference propagaton tme ntervals along the lnes between two adjacent slaves must be determned. Ths can be done, for nstance, by carryng out smulatons n Electro Magnetc Transent Program (EMTP) envronment, or by computng the propagaton speed of sgnals on the bass of the theoretcal model of the lnes. Slave unts should be nstalled n each network termnal, n each node, and at ponts where the number of phases eventually vares. The procedure proposed for locatng the source of transents reles on the obvous concept that the more dstant s a slave from the source, the greater s the tme stamp value t regsters n correspondence of the dsturbance arrval. Let t mj be the tme stamp of the startng nstant of the transent voltage measured by the generc mth slave (1 m M) on the jth phase (j = a, b, c n the case of a three-phase network). Moreover, T njref and T njmeas denote respectvely the reference and measured propagaton tme ntervals on the jth phase of the nth lne secton (1 n M-1). The two adjacent slaves at the ends of the consdered lne are separated by the dstance D nj that s the nomnal length of the nth lne secton. In deal condtons, when the nth secton s safe, T njmeas should be dentcal to T njref. Despte, due to the effects of the dfference between study hypothess and real condtons whch lead to a non-perfect estmaton of some parameters for all the operatng condtons, T njmeas can dffer from the relevant T njref. Anyway, the former parameter s always lower than the latter one when the source of the transent s located wthn the consdered couple of ponts. The algorthm runnng on the master unt processes the data smultaneously sent by the slaves,.e. the tmestamps. For each phase and each couple of adjacent ponts, the followng quantty s determned:

97 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS T T nj ref meas nj = (3.47) T nj ref nj whch ranges from 1 to 0 as the transent source moves from the mdst of the nth lne up to outsde the lne tself. Then, ( nj ) max.e. the maxmum for nj s found, to dentfy whch phase and above all whch lne secton s affected by the fault. Once the faulted lne s detected, ( nj ) max allows also determnng the dstance d of the transent source from the nearest slave: d D nj = ( nj ) max (3.48) 2 D nj d m 0 m+1 t mj t 0 t m+1j Fgure nth lne, connectng node m to node m+1, D nj long To explan the fnal expresson used n the locaton algorthm, Fgure 3-29 can be consdered: the secton between two generc adjacent slaves separated by the dstance D nj. Let us assume that the transent event starts n pont 0 at tme nstant t=t 0 and s detected by the slaves m and m+1 at nstants t mj1 and t m+1j, respectvely. If T njref s the propagaton tme nterval relevant to the consdered secton, thus the reference propagaton speed v njref s: Dnj v njref = (3.49) T njref The dstances of the transent source from the consdered slaves can be expressed as follows: ( tmj t 0 ) v ( t ) d = v (3.50) D nj njref d = + (3.51) njref m 1 j t 0 d s the dstance between the pont 0 and the closest slave. By subtractng (3.51) to (3.50) we get: D ) (3.52) nj 2 d Dnj = ( tmj tm+ 1 j Tnjref By denotng T njmeas = (t mj t m+1j ) and then substtutng (3.47) n (3.52), the dstance d s fnally determned:

98 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS Dnj Tnjref Tnjmeas Dnj d = = ( nj ) 2 T 2 max (3.53) njref Some consderaton about the proposed approach can be drawn. The EMTP smulaton campagn on both the dstrbuton feeders that wll be presented n the next Chapter proves that the method descrbed above features very well the fault locaton n all the smulaton cases, leadng to low-based results. Such good performance of the procedure are manly due to: a) the large number of measurement ponts; b) the senstvty of EDB devces nstalled n the slave unts, whose threshold value has a lower lmt dependng on the qualty of voltage waveforms just to avod fake detectons n steady state condtons. Despte, by the economcal pont of vew, the method would have a weak appeal because of ts relatvely large captal and nstallaton costs. As a matter of fact, n the confguraton descrbed above, the method requres a slave unt at the head and at the end of each lne. The cost of each remote staton s n the order of 6,000. Amng at decreasng the global cost of the dstrbuted measurement system, the fault locaton algorthm runnng on the master unt s proposed to be modfed so that the number of slave unts s reduced of around 50%, n fact f M s the number of remote statons nstalled to montor a dstrbuton network, the dfferent approach needs only (M/2 + 1) slaves

99 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS 3.6. Fault locaton method ntegratng the dstrbuted measurement system and wavelet analyss The alternatve verson of the fault locaton method reles on a proper ntegraton between a dstrbuted measurement system performng the procedure descrbed n prevous secton and a sgnal processng technque based on the Wavelet transform. In partcular, the number of remote unts decreases because for each lateral lne of the network sngleended measurements are used nstead of double-ended ones. In ths connecton, for a typcal radal dstrbuton system, the slave unts are placed at the head and at the end of the man power feeder and at the connectons wth laterals gettng to the load busses. If the prevous procedure were adopted for fault locaton on the whole dstrbuton network, also remote statons nstalled n all the load busses (.e. at the end of each lateral) were needed. The lack of nformaton on the propagaton of the dsturbance no longer avalable at the latter measurement ponts s compensated n the new method by the analyss of the voltage waveforms acqured n the nodes along the man feeder. In ths connecton, the Dscrete Wavelet Transform (DWT), whch s the dgtally performed verson of the Contnuous Wavelet Transform, s mplemented. The travellng-waves theory s exploted to estmate the poston of a fault along the laterals. In practce, the sectons of the man feeder are stll montored by the double-ended technque, whereas the lateral lnes are montored n a sngle-ended way. To descrbe some detals of the method let us refer to a typcal dstrbuton network montored by the fault locaton system. The topology of the power system s shown n Fgure It conssts n a 10-km long man feeder (lnes L1, L2 and L5) and n two laterals: L4 (2-km long) and L3 (1-km long). In the system three-phase overhead lnes are used. The radal dstrbuton network s fed by a 150/20 kv substaton G, where the transformer T s of the Y g /d connecton type. All the load busses are equvalent to threephase balanced mpedances. Each load s connected to the relevant lateral through a 20/0.4 kv transformer. Four slave unts of the dstrbuted measurement system, denoted by S1, S2, S3 and S4, are placed n correspondence of the black bullets n Fgure As any type of short crcut occurs n one of the lnes, the transent waveforms and the relevant tme stamps are acqured by the four slave statons and sent to the master unt, whch detects the nearest slave to the fault poston, updates the value of the propagaton speed on the bass of the modal voltages computed for the selected node, and then obtans the value of delta for the sectons L1, L2 and L5. The delta values are compared wth a threshold, n order to check f the faulted lne belongs to the man feeder and eventually dentfy the secton affected by the fault. The value of delta greater than the

100 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS threshold allows to detect the faulted secton of feeder, hence the master unt can mplement (3.48) to compute the poston of the transent source along that lne. Fault locaton Remote staton L4: 2 km Load Bus 2 B G T S 2 L1: 2 km L2: 3 km L5: 5 km S 4 S 1 S 3 A Load Bus 1 L3: 1 km Load Bus 3 Fgure Dstrbuton network (three phase overhead lnes) montored by the measurement system. For example ths would be the case of a short crcut n A n Fgure on feeder sectons > Threshold? T faulted secton on feeder: seek max F faulted lne seek t mn: nearest slave to fault D d = 2 faulted lateral DWT analyss of three-phase voltages Travellng waves approach Dstance slave - fault Fgure Block dagram of the ntegrated fault locaton method. If the threshold s larger than the delta values relevant to all sectons L1, L2, L5, the man feeder of the network s assumed to be not affected by faults, and thus the algorthm dentfes the faulted lne as the lateral whose head s montored by the slave unt that recorded the mnmum tme stamp. The dstance of the fault pont from the selected measurement unt s computed by analyzng the three phase voltage sgnals acqured at the head of the selected lne by means of the DWT. Ths procedure would be appled to a fault occurrng n pont B n Fgure As Mother wavelet the Daubeches 4 has been chosen, because of the hgh correlaton degree between ts pattern and the typcal oscllatons of the dumped transents developng n the network as the fault occurs. In order to choose the most useful mother

101 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS wavelet to the purpose, prelmnary smulatons tests have been carred out by consderng a sngle three phase lne fed by one end and loaded by the other end, affected by a short crcut to ground. The lne voltages at the source pont were analysed by applyng the DWT based on dfferent mother wavelets. The set of mother wavelet to be used was selected accordng to ther waveshape compared to the oscllatons of a transent dsturbance. Also the value of the fault resstance was vared. At the end of such analyss the mother wavelet leadng to the hghest number of correct locatons was adopted. The equvalent samplng frequency of the dstrbuted measurement system (10 MSa/s) leads to a frequency sub-band for the frst DWT level of Detals n the range [2.5, 5] MHz. The man spectrum content of transents generated by short crcuts s known to be largely below 1 MHz, n fact the energy content assocated to the frst level of coeffcents s always lower than other nferor frequency sub-bands. Desptes, amng at computng the dstance of the transent source from the slave on the bass of the travellng waves theory, the detal coeffcents of the frst level of decomposton are analyzed, because hgh coeffcents are found only when the voltage transent presents very fast rsng and fallng fronts, and thus a sngle spke can be dentfed. The proposed strategy to detect when an event occurs s enforced by a huge number of expermental tests n whch the detals belongng to the 5 hgher decomposton levels of the sgnals acqured at the head of a lne were analyzed. Both the ampltude and the energy content of each detals level were taken nto account as nformaton on whch the poston of the fault along the lne could be estmated; the type of fault, ts resstance as well as poston were systematcally vared n order to understand whch decomposton level could gve correct results n most cases. Fnally, t should be remnded that the best tme resoluton s obtaned at the frst decomposton level. By montorng the coeffcents sequence (and thus the oscllatons belongng to the transent waveform) the algorthm: ) seeks the frst pulse front (no matter f rsng or fallng) assocated wth the travellng wave propagatng from the fault pont to the head of the lne; ) selects a sequence of detals startng wth the frst pulse whose length corresponds to the propagaton tme of the surge from the head to the end of the lne and back; ) n ths tme wndow seeks the second spke, assumed as the travellng wave reflected once from the measurement pont to the fault, and then back from the fault poston to the slave. From a computatonal pont of vew, operatons ) and ) become easer f the squares of the detals are used. Actually n dong ths the nformaton on the sgn of each transent

102 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS surge s lost; by consderng that t s due to the value of reflecton and refracton coeffcents at the fault and load ponts t could sound lke loosng useful data. Despte, the equvalent mpedance of the fault and the power absorbed by the load bus at the occurrence of the fault tself cannot be known, hence no general relatonshp let further nformaton on the fault condton be found by analyzng the surges sgn. A crtcal aspect of the analyss descrbed for estmatng the dstance from the fault pont to the measurement unt s that the order by whch the surges get to the measurement pont depends on the fault poston along the montored lne: the concept can be easly shown referrng for example to Fgure The oscllaton regstered n pont A for t=3t 1 due to the surge reflected by the fault pont s the second event only f 0 < x < 2l/3. For faults dstant more than 2l/3 from pont A the pulse (t = T 1 + 2T 2 ) due to the surge reflected n B and refracted n F gets to pont A before the above one. The dstance x s not known, obvously, and hence such condton can be msleadng. Moreover, the equvalent resstance of the fault n F has a strong nfluence on the ampltude of the refracted surges respect to the reflected ones, but also the fault resstance cannot be known a pror. Tryng to overcome ths ssue, the refracton coeffcent of the fault pont s estmated by checkng the ampltude of the coeffcents n correspondence of the tme t = (3T 1 + 2T 2 ) n Fgure 3-17,.e. the tme when the frst surge seen n pont A (t = T 1 ) gets back there after travellng along the lne tll the reflecton pont B. If a sgnfcantly hgh coeffcent s found the refracton coeffcent of the fault pont (and hence ts resstance) s assumed to be hgh. As the couple of tme nstants N 1 T s and N 2 T s (T s = 1/f s ) relevant to the above events are known, the dstance d l of the short crcut pont from the head of the selected lateral s computed by means of the followng relatonshp: D ( N N ) va l = Ts 1 2 (3.54) where v α s the propagaton speed relevant to the α mode

103 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS References [1] IEEE std , Recommended practce for montorng electrc power qualty, The IEEE, Pscataway (USA), Nov. 1995; [2] A. Haar, Zur theore der orthogonalen funktonen-system, Math. Ann., vol. 69, 1910, pp [3] A. Grossmann, J. Morlet, Decomposton of Hardy functons nto square ntegrable wavelet of constant shape, SIAM J. Math. Anal., vol. 15, 1984, pp [4] I. Daubeches, Orthonormal bases of compactly supported wavelets, Comm. Pure & Appl. Math., vol. 41, 1988, pp ,. [5] S. G. Mallat, A theory for multresoluton sgnal decomposton: the wavelet representaton, IEEE Trans. on Pattern Analyss and Machne Intellgence, vol. 11, no. 7, 1989, pp ,. [6] C. Hel, D. Walnut, Contnuous and dscrete wavelet transform, SIAM Revew, vol. 31, 1989, pp [7] G. Strang, Wavelets and dlaton equatons: a bref ntroducton, SIAM Revew, vol. 31, 1989, pp [8] P. F. Rbero, Wavelet transform: an advanced tool for analyzng non-statonary harmonc dstortons n power system, Proc. IEEE ICHPS VI, Bologna, Italy, 1994, pp [9] W.B. Rchardson Jr., Applyng wavelets to mammograms, IEEE Engneerng n Medcne and Bology Magazne, vol. 14, n. 5, 1995, pp [10] S. G. Mallat, Multfrequency channel decompostons of mages and wavelet models, IEEE Trans. on Acoustcs, Speech, and Sgnal Processng, vol. 37, n. 12, 1989, pp [11] L. Angrsan, P. Daponte, M. D Apuzzo, A. Testa, A new wavelet transform based procedure for electrcal power qualty analyss, Proc. ICHQP 96, Las Vegas, USA, 1996, pp [12] L. Khadra, M. Matalgah, B. el-asr, S. Mawagdeh, The wavelet transform and ts applcatons to phonocardogram sgnal analyss., Med Inform, vol. 16, n.3, 1991, pp [13] J. G. Tet, Jr., H. N. Krtkos, SAR ocean mage representaton usng wavelets, IEEE Trans. on Geoscence and Remote Sensng, vol. 30, n. 5, 1992, pp [14] G. Strang, T. Nguyen, Wavelets and flter banks, Wellesley-Cambrdge Press, Wellesley, USA, [15] O.L. Wlls, A revew of fault locatng technques n medum-voltage power cable, n Proc Petroleum and Chemcal Industry Conf., pp [16] Cred Report, Fault Management n electrcal dstrbuton system, Tech. Rep. CIRED workng group WG03, Dc [17] Y. Tang, H.F. Wang, R.K. Aggarwal, and A.T. Johns, Fault ndcators n transmsson and dstrbuton systems, n Proc Internatonal Electrc Utlty Deregulaton and Restructurng and Power Technologes Conf., pp [18] IEEE gude for testng faulted crcut ndcators, IEEE Standard , Apr [19] A. A. Grgs, C.M. Fallon, and D.L. Lubkeman, A fault locaton technque for rural dstrbuton feeders, IEEE Trans. Industry Applcatons, vol. 29, pp , Nov.-Dec

104 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS [20] J. Zhu, D.L. Lubkeman, and A.A. Grgs, Automated fault locaton and dagnoss on electrc power dstrbuton feeders, IEEE Trans. Power Delvery, vol. 12, pp , Apr [21] R.K. Aggarwal, Y. Aslan, and A.T. Johns, New concept n fault locaton for overhead dstrbuton systems usng supermposed components, Proc. Inst. Elect. Eng.- Gener. Transm. Dstrb., vol. 144, pp , May [22] P.M. Van Orsouwf and F. Provoost. Fault localsaton n an MV dstrbuton network, n Proc CIRED 17th Internatonal Electrcty Dstrbuton Conf. [23] M. El-Ham, L.L La, D.J. Daruvala, and A.T. Johns, A new travelng-wave based scheme for fault detecton on overhead power dstrbuton feeders, IEEE Trans. Power Delvery, vol. 7, pp , Oct [24] Z.Q. Bo, G. Weller, F. Jang, and Q.X. Yang, Applcaton of GPS based fault locaton scheme for dstrbuton system, n Proc POWERCON 98 Internatonal Power System Technology Conf., Bejng, Chna, 1998, vol. 1, pp [25] D. W. P Thomas, R. J. O. Carvalo, E. Perera, Fault Locaton n Dstrbuton Systems based on Travellng Waves, 2003 IEEE PowerTech Conference, Bologna, Italy, June [26] F.H. Magnago and A. Abur, A new fault locaton technque for radal dstrbuton systems based on hgh frequency sgnals, n Proc 1999 IEEE Power Engneerng Socety Summer Meetng Conf., Edmonton, Alta, Canada, 1999, vol. 1, pp [27] Z.Q. Bo, G. Weller, M. A. Redfern, accurate fault locaton technque for dstrbuton system usng fault-generated hgh-frequency transent voltage sgnals, IEE Proc. On Gener. Transm. Dstrb., Vol. 146, No. 1, January [28] H. Hzam, P.A. Crossley, P.F. Gale, and G. Bryson, Fault secton dentfcaton and locaton on a dstrbuton feeder usng travelng waves, n Proc IEEE Power Engneerng Socety Summer Meetng, vol. 3, pp [29] A. Borghett, S. Cors, C.A. Nucc, M. Paolone, L. Peretto and R. Tnarell: On the use of contnuous-wavelet transform for fault locaton n dstrbuton power systems, Internatonal Journal of Electrcal Power & Energy Systems, vol. 28, n. 9, November 2006, pp ; [30] H. Yuan-Yh, F.C. Lu, Y. Chen, J.P. Lu, J.T. Ln, P.H.S. Yu, and R.R.T. Kuo, An expert system for locatng dstrbuton system faults, IEEE Trans. Power Delvery, vol. 6, pp , Jan [31] P. Jarventausta, P. Verho, M. Karenlamp, and J. Partanen, AI-based methods n practcal fault locaton of medum voltage dstrbuton feeders, n Proc ISAP 96 Internatonal Intellgent Systems Applcatons to Power Systems Conf., pp [32] J. A. Momoh, L.G. Das, and D.N. Lard, An mplementaton of a hybrd ntellgent tool for dstrbuton system fault dagnoss, n Proc IEEE Transmsson and Dstrbuton Conf., pp [33] S. Hännnen, M. Lehtonen, and U. Pulkknen. A probablstc method for detecton and locaton of very hgh resstve earth faults, Electrc Power Syst. Res., vol. 54, pp , Jun

105 3. ELECTROMAGNETIC TRANSIENTS IN POWER DISTRIBUTION NETWORKS [34] L. Peretto, P. Rnald, R. Sasdell, R. Tnarell, A system for the measurement of the startng nstant of mpulsve transents, Proc. of the IEEE IMTC/2004, Como, Italy, May 2004, vol.2, pp

106

107 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT 4. Analyss of the performance featured by the fault locaton method based on dstrbuted smultaneous measurements - smulatons of dstrbuton networks n EMTP-RV envronment 4.1. Implementaton of an IEEE Std. test network affected by bolt faults IEEE proposed n [1] four models of network to be used by researchers nvolved n power qualty analyss to compare the results of dfferent nvestgaton methodologes. Among these models, the IEEE 34-Node Test Feeder, shown n Fgure 4-1, was chosen to test the performance of the locaton methodology for transent sources n dstrbuton systems. Such an overhead network s characterzed by a radal dstrbuton of three-phase lnes where a part of sngle-phase laterals can be found; the entre network s suppled by a substaton transformer havng the followng specfcatons: 2500-kVA rated power, 345- kv/24.9-kv voltage rato, -y grounded connecton, per-unt equvalent mpedance equal to: (0.01+j0.08). The three-phase short crcut power, on the 345-kV sde, s 1800 MVA at an angle of 85 degrees. Both spot and dstrbuted Y connected loads havng constant actve and reactve power are nstalled n the network. In the document [1] the actve and reactve nomnal powers per phase are gven for all the spot and dstrbuted loads. Fnally, an n-lne autotransformer convertng the feeder voltage from 24.9 kv to 4.16 kv s used to supply the termnal 890. The network n Fgure 4-1 has been mplemented n EMTP envronment and the transent voltages too. Table 4-1 reports the values of D nj and T njref for the network n Fgure 4-1. The values of T njref have been determned by smulatng the njecton of a voltage pulse n pont 800. The proposed locaton procedure s based on the measurement of tme nstants; hence, the smulatons have been performed n tme doman. The ntal condtons for each smulaton run are gven by the 50 Hz steady state obtaned by EMTP as load-flow soluton. In ths way the entre duraton of the smulaton s useful to test the propagaton of the dsturbance snce the fault occurs. The smulaton tme step has been set 10 ns to grant a tme resoluton suffcently hgher than the accuracy assocated wth the real employed nstrumentaton. For the overhead lnes of the network, the Constant Parameters Lne Model has been adopted durng smulatons. Such a choce can sound contradctory respect to the lnes theory shortly reported n Secton 3.2, accordng to whch an exhaustve model of the electrc lne should take nto account the dependence on frequency of the equvalent

108 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT dstrbuted parameters. Indeed, the characterstc mpedance of the lne would be nfluenced by the sgnal frequency content. Actually, the lne characterstc n functon of frequency s almost flat, and hence usually the operator assumes constant values for the model parameters. In dong ths, each parameter value has to be chosen n correspondence of a frequency representatve of the phenomenon to be smulated. In partcular, for the approxmaton of the model wth the CP lnes, n the case of transents generated by faults the lne parameters can be computed at 100 khz, because the typcal frequency bandwdth assocated wth the frst oscllatons belongs to the range 10 khz 500 khz. In order to verfy the correctness of neglectng the frequency dependence of the dstrbuted parameters n the model, at frst smulatons have been run by consderng the same condton and locaton of transent source n the network under test, and by loadng n the former case the FD-lne, n the latter case the CP-lne model for all the sectons. The results of the fault detecton and locaton procedure were dentcal n the two cases. Montored node V source Phase-to-phase fault Phase-to-ground fault Sngle-phase lne Three-phase lne Three-phase capactor bank Fgure 4-1 Dagram of the IEEE 34-node Test Feeder (not n scale) It should be noted that the MV dstrbuton networks used for tests, as well as actual radal networks, are characterzed by qute short laterals, less than 10 km long. The dependence on frequency of the lne model used durng transent smulatons could sgnfcantly affect the results for lnes length much hgher than a few km. Moreover, the

109 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT nfluence of the frequency content of the sgnals propagatng along the lnes could lead to dfferences n the values of reflecton or refracton coeffcents n the network nodes respect to the case of CP lnes, and thus lead to a dfferent transent pattern obtaned as superposton of drect, reflected and refracted waves. Desptes, the method proposed for the locaton of transent sources uses nformaton on the propagaton of the travellng waves strctly related to the frst drect surge, and no analyss of the hgh frequency sgnal s carred out, hence t seems reasonable that the measured propagaton tmes are not sgnfcantly nfluenced by the lnes model. As a matter of fact, the network consdered for tests s homogeneous, n the sense that the only parameter varyng from one lne secton to any other one s length. Ths means that dfferent values of the lnes parameters turn nto a dfferent value of the propagaton speed of travellng waves, but the order by whch the slave unts can detect the arrval of the dsturbance remans the same. Table 4-1 lnes length and reference transmsson tme ntervals relevant to the Network represented n Fgure 4-1 Slave Couples D na [m] T n a,ref [µs] D nb [m] T nb,ref [µs] D nc [m] T nc,ref [µs] Dfferent events gvng rse to damped oscllatory transent voltages have been smulated:. short crcuts between two phases (n dfferent ponts marked by stars n Fgure 4-1;. phase-to-ground short crcuts (n ponts marked by flashes);. nserton of three-phase capactor banks n nodes marked by arrows (100-kVAR rated power n node 844 and 150-kVAR n node 846, respectvely). In the cases ) and ) the capactor banks were dsconnected

110 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT No slave unt has been set n 890 because t s characterzed by a dfferent nomnal voltage. In practcal applcatons t should be montored by another measurement system, gven that t belongs to a dfferent network. The results have been obtaned by processng n Matlab envronment the voltage sgnals provded by EMTP smulatons. In all the montored ponts three-phase voltage probes have been nstalled. In ths frst tests set for the valdaton of the proposed method, Matlab has been used to smulate the operaton of the whole measurement system n terms of both slave and master unts, thus frst determnng the tme stamps, then locatng the transent source on the bass of tme data. At the end of each fault smulaton carred out n EMTP-RV envronment, the waveforms of the voltage sgnals acqured by the probes n the montored ponts of the test network are mported n the Matlab Workspace. The followng elaboratng procedure s repeated for each one of the three phases: A functon emulatng the Event Detecton Block s appled to the mported data so that the tmestamps assocated to the begnnng of the transent dsturbance are found for all the montored ponts. The functon mplemented compares the dfference between each sample and the prevous one of the nput waveform to a threshold value and gves as output the tme reference correspondng to the frst overcome of the threshold. The tmestamps are saved nto an array. Every secton of lne belongng to the network s assocated to a couple of montorng ponts correspondng to ts ends: the mnmum of the two tmestamp s found. Then, the measured propagaton tme s computed as the dfference between the two tmestamps. On the bass of the nomnal lne length, also the reference propagaton tme s computed for each lne secton. The parameter delta s computed for all the lnes, then the maxmum delta s found. The lne assocated to the maxmum delta s declared as faulted; ts length and the mnmum of the two tmestamps at ts ends are used to locate the fault. To obtan the hghest accuracy n the tme reference and hence n the locaton of the fault, the absolute maxmum value between the three relatve ( ) max found for the phases s used to compute the dstance d. Moreover, the phase conductor correspondng to the absolute maxmum s for sure affected by the fault, because the begnnng of the transent on ths voltage occurs earler than on the safe phases. Anyway, the electromagnetc couplng between the conductors of the three phases belongng to the same overhead lne leads to fnd the same transent dsturbance on all the lne-to-ground voltages, so that no nformaton can be gven on the number of phases nvolved n the fault

111 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Expermental Results Lne-to-lne short crcut An deal swtch, closed 1.5 ms after the smulaton startng, has been employed to smulate a short crcut between two phases. The postons of the fault, along wth the phases nvolved, have been changed durng the dfferent smulatons. Fgure 4-2 s the plot of the voltage sgnal at node 858 caused by a short crcut between phases a and c n 828. Fgure 4-2 Transent pattern regstered n node 858 when the lne-to-lne short crcut occurs n pont 828 By way of example, Table 4-2 reports the values of nj estmated for all the portons of the network under test, n the case of two short-crcut locatons. In both ths table and the followng ones, the values of ( nj ) max are n bold characters. The lne between the nodes 824 and 854 s correctly dentfed as faulted when the fault occurs n 828 between phases a and c (case #1). Moreover, the secton s correctly marked as faulted when the fault occurs n 806 between phases a and b (case #2). The values of d computed accordng to (3.48) are 257 m for case #1 and 1312 m for case #2. The actual dstances are 257 m and 1314 m, respectvely. The proposed procedure shows dentcal good performance when short-crcuts are smulated between both phases b-c n 830 and a-b n 860, thus the results are not reported

112 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Table 4-2 nj estmated for all the portons of the network n the case of two short-crcut locatons Fault n 828 (a-c phases) Fault n 806 (a-b phases) Nodes couples na ( 10-2 ) nb ( 10-2 ) nc ( 10-2 ) na ( 10-2 ) nb ( 10-2 ) nc ( 10-2 ) Lne-to-ground short crcut An deal swtch closed 1.5 ms after the smulaton startng, has been employed to smulate a phase-to-ground short crcut. The poston of the fault has been changed n the smulatons. Fgure 4-3 shows the damped transent voltage caused n node 854, phase a, by the fault n 862. Table 4-3 reports the values of nj for the short crcuts between phase a and ground n pont 820 (case #3) and also between phase a and ground n 862 (case #4). The sectons marked as faulted are correctly dentfed between slave unts and , respectvely. As far as the dstance of the fault from the nearest slave s concerned, the applcaton of (3.48) provdes 4188 m for case #3 and 85 m for case #4; these computed dstances are equal to the actual ones

113 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Table 4-3 nj regstered for the phase a to-ground short crcuts n ponts 820 (case #3) and 862 (case #4) Fault n 820 (a-gnd) Fault n 862 (a-gnd) Nodes couples na ( 10-2 ) nb ( 10-2 ) nc ( 10-2 ) na ( 10-2 ) nb ( 10-2 ) nc ( 10-2 ) Inserton of a capactor bank An deal swtch (closed 1.5 ms after the begnnng of the smulaton) has been used to connect each capactor bank to the relevant node. Table 4-4 shows the values of nj obtaned as a consequence of the nserton of a three-phase balanced capactor bank wth reactve power equal to 300 KVAr n node 844 (case #5) and 450 kvar n node 846 (case #6). In both cases, the porton of network nto whch the sources of the transents are located s correctly dentfed as The computed dstances are: 499 m from node 834 (case #5) and 162 m from termnal 848 (case #6), whereas the actual ones are 497 m and 162 m, respectvely. The capactor banks n ponts 844 and 846 have been assumed normally not connected on the bass of frst tests. In fact, the fault locaton method s not relable f a fault occurs n any pont of the network when the capactor banks are already connected n the relevant nodes. In such condtons the equvalent mpedance of the lne s very much dfferent wth respect to the other overhead lnes, and the capactors represent a sort of short crcut for the hgh-frequency components of the sgnals due to the transent event

114 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT In practce, even f the lne s safe, the measured propagaton tme obtaned on the bass of the propagaton of the transent from one end to the other dffers from the reference propagaton tme of a quantty that s not comparable to the usual non-zero dfferences due to the ntrnsc uncertanty contrbuton of the method. In ths case the maxmum value of delta cannot be assocated any longer wth the faulted lne, but most of the tmes corresponds to the lne where capactor banks are connected. Ths ssue does not seem to be solved by evaluatng how much the presence of capactors nfluences the propagaton of the travellng waves, because ther contrbuton s not constant, as well as the power of the banks n real operatng condtons. Fgure 4-3 Transent pattern regstered n node 854 when the lne-to-ground short crcut occurs n pont 862 Fgure 4-4 Transent pattern regstered n node 854 when the capactor bank s connected n pont

115 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Table 4-4 nj regstered for the connecton of capactors n ponts 844 (case #3) and 846 (case #4) Capactor bank n 844 Capactor bank n 846 Nodes couples na ( 10-2 ) nb ( 10-2 ) nc ( 10-2 ) na ( 10-2 ) nb ( 10-2 ) nc ( 10-2 ) In order to obtan correct results for the poston of the fault wthn the test network, the number of slave statons should be ncreased and two more measurement ponts should be added to the system: just the nodes where capactor banks are connected to the lne. Ths s the smplest but even the most robust way to evaluate correctly how the transent propagates along the lne , montorng three separate sub-sectons. Some consderatons can be drawn from the results presented n Tables 4-2, 4-3, 4-4. The proposed method correctly dentfes the lateral or the secton of lne where the transent occurs and computes wth good accuracy the dstance between the source and the nearest slave. However, non-zero values of nj are found also for non-faulted sectons. Ths s reasonably due to the superposton of drect and reflected waves whch, case by case, modfy the voltage waveforms n the nodes. Therefore, errors may arse n the measurement of T njmeas and, hence, n the relevant nj. Snce the effect of reflected and refracted waves depends on the transent source locaton, the hypothess s strengthened by consderng the number of zero values of nj. The results relevant to cases #2 and #3, whch refer to faults located close to pont 800, feature the hghest occurrence of zero values. Pont 800, next to the HV/MV staton, corresponds to the pont where the pulse used to determne T njref was njected. The effect of the reflected waves on T njref and T njmeas s therefore almost the same; hence, no apprecable dfference between T njmeas and T njref

116 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT s found. The mpact of ths error turns nto a reduced senstvty when computng the dstance d. Indeed, when the transent source s located near a slave, the value of the relevant nj could not be ( nj ) max ; n such a case the method would not provde correct nformaton. In Tables 4-2, 4-3 and 4-4 the largest ncorrect value of nj s n the order of 10-2, whch corresponds to a resoluton d mn = 0.5% D nj. It s worthwhle emphaszng that such resoluton depends on the dstance between the couple of slave unts, whereas the uncertanty affectng the measurement of d s a constant value. The longest secton of the network (.e. the one between pont 808 and node 816) leads to d mn = 100 m. Ths value s lower than the extended standard uncertanty affectng the fault locaton due to the only contrbuton of the GPS-devce f a coverage factor equal to 3 s taken, whch s 180 m. In order to smulate actual stuatons, also the presence of nose affectng the sgnals of the power system voltage to be processed should be consdered. Nose conssts of any unwanted dstorton of the power sgnal that cannot be classfed as harmonc dstorton or transent [2]. It has broadband spectral content lower than 200 khz and may arse from the operaton of power electronc devces, control crcuts, arcng equpment, loads wth sold state rectfers and swtchng power supples. The frequency range and magntude level of nose depend on the source, whch produces the nose and the system characterstcs; a typcal magntude of nose s less than 1% of the voltage magntude, [2]. Moreover, the voltage sgnals are also affected by whte nose arsng from the measurement hardware. For the above reasons, whte nose has been added to the system voltages and new smulatons have been run for cases #1, #3, and #6. The nose has been fltered n order to cancel the components wth frequency hgher than 200 khz; a nose ampltude close to 200 V peak to peak has been assumed. Of course, the voltage threshold value n the algorthm smulatng the operaton of the EDB has been ncreased. In all the three cases the method correctly both dentfes the power system porton where the source of the transent s located and determnes the value of d. In partcular, the smulaton results for cases #1 and #3 are just the same as those obtaned wthout nose superposed to the sgnal. As for case #3, values of d have been found that are slghtly dfferent from those acheved wthout nose as far as the dfferent phases are consdered; however, the dfference s n the order of 1.2%. Ths leads to conclude that the measurement system performs very well even n the presence of nose. The analyss of the expermental results allows to state that the approach performs very satsfactorly. Gven that other factors usually affect the operaton of a real network, smulatons to test the performance of the proposed method n dfferent condtons have been run

117 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT 4.2. Implementaton of a dstrbuton network operatng n actual condtons varaton of fault parameters Improvements n the procedure Test results reported n former Secton show that n case of phase-to-phase and phaseto-ground short crcuts the prevous verson of the algorthm correctly detects the faulted lne secton and locates the fault pont wth good accuracy. However, durng the above smulaton tests some more nformaton on the system and fault condtons came out by observng the obtaned data set. Amng at mprovng the system performance n terms of both accuracy and relablty of results n general condtons, a couple of operatons have been added to the procedure. Frst of all, the master unt seeks for the mnmum tme stamp value (t mj ) mn before analyzng any data sent by the slaves. (t mj ) mn dentfes the slave nearest to the fault, so the locaton procedure can be appled only to a lmted network area. The master selects the group of lne sectons havng one common end n the node assocated to the (t mj ) mn. Smulaton tests reported n the followng showed that ths smple operaton strongly reduces both the computaton tme and above all the occurrence of wrong results. In fact, seekng for the (t mj ) mn allows to detect whch one of the phases s certanly affected by the fault. Despte, f grounded short crcuts characterzed by a hgh value of the fault resstance are consdered, the dumped oscllatons nduced on the voltages are very small. The EDB nstalled n the slave unts cannot detect the correct tmestamp at the begnnng of the transent voltage on each phase and ths fact can turn nto a wrong operaton of the locaton procedure, often leadng to dfferent results on the three phases for a fxed poston. In such cases the locaton of the fault s assumed as a correct result only f the three postons obtaned by the data relevant to each sngle phase are coherent wth each other. Moreover, losses on actual electrc conductors dump the oscllatng dstorton as t propagates farer and farer from the transent source, thus compromsng the fault locaton system operaton. Indeed, the ampltude of the transent can be lower than the threshold value of the EDB n slaves far from the dsturbance source, hence some of them could not detect the event. The algorthm has been mproved n order to tackle ths ssue: the faulted lne s dentfed by comparng the values of delta no longer relevant to all the lnes, but only to the lnes ncluded between a couple of slaves detectng the event. In practce the mnmum number of tmestamps needed for the locaton of the fault s two, and n such a case the two slaves detectng the event wll be the ones montorng the ends of the faulted lne

118 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Fnally, the three phase-sgnals regstered by the slave unt nearest to the fault are consdered: the transformaton from phase to modal voltages s appled and the zero mode component s analyzed. In ths way grounded faults can be dstngushed from ungrounded ones on the bass of the magntude of the zero-mode voltage, and a more correct value of the propagaton speed can be assumed n both cases for v njref. In fact, for grounded short crcuts the mode relevant to the couplng wth ground prevals, whch s characterzed by a slghtly lower propagaton speed of the travellng wave respect to the case of phase-to-phase faults. In Fgure 4-5 the block dagram of the updated procedure s represented. Topology and man characterstcs of the network T nj ref Data sent by slaves as a fault occurs: tmestamps t mj waveforms acqured by DAQ boards Master unt algorthm computes (t mj ) mn T nj meas nj Nearest slave to fault locaton Selecton of area where seekng ( nj )max Lne affected by the fault Analyss of voltage waveforms sent by the nearest slave: grounded/ungrounded fault Dstance d of the fault poston from the nearest slave Fgure 4-5 upgraded verson of the procedure mplemented by the master unt Test condtons: non deal faults and network elements In the above smulatons of the MV network realstc fault parameters were not yet taken nto account because faults were modelled as deal short crcuts. Actually, the ampltude of the transent generated by the fault depends on the values of several parameters at the occurrence of the event,.e. the poston of the fault pont, the type of fault, the magntude of the equvalent fault resstance R f and the nstantaneous voltage value between the faulted conductors. To show the strong nfluence of the latter couple of fault parameters on the magntude of the event, Fgure 4-6 can be consdered. Three acqustons of the lne-to-ground voltages are reported, relevant to the same slave node and the same poston of the phase a-to-ground short crcut, to underlne the mpact of R f and V f on the network

119 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Smulatons have been run n EMTP-RV envronment [3] to test the ablty of the measurement system to detect and locate faults affectng a typcal radal dstrbuton network operatng n real condtons. Most frequent knd of faults have been smulated: phase-to-ground, phase-to-phase short crcuts and lghtnng events; all of them generate dumped oscllatory transents. Moreover, nose and harmonc dstorton of the voltage have been consdered to verfy f such crtcal parameters could nfluence the detecton method performance. As above explaned, nose could lower the senstvty of the measurement system leadng to wrong nformaton. In order to avod fake event detectons the threshold value of each EDB has been ncreased tll 2% of the power system component of u VVT. u = voltage on the faulted phase when the short crcut occurs. u = 15.3 kv R f = 0 2 x V[V] Va Vb Vc t[s] x x 104 u u = 15.3 kv R f = 2.85 kω Va Vb Vc 0 V[V] t t[s] x x Va Vb Vc u = 1.2 kv R f = 0 V[V] t[s] x 10-3 Fgure 4-6 patterns of the transent dsturbances affectng lne voltages n case of dfferent lne-toground short crcut condtons

120 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT The proposed method needs to know data relevant to the topology of the montored network n order to feature correct locaton of the faults. In usual operatng condtons a dstrbuton network has not a constant and fxed topology, therefore smulatons have been run takng nto account manoeuvres that may occur n a real network. Varatons of loads power, dsconnecton of lnes or loads have been mplemented to test whether confguraton changes of the network can lead to wrong results. The measurement system has also been tested n the presence of more than one of the above parameters Models of Power dstrbuton and Measurement Systems In prevous smulatons, carred out to test the frst verson of the method, both the measurement chan representng the remote unt and the locaton procedure were mplemented by usng Matlab scrpts. In ths case most of the hardware nstalled n the slaves has been mplemented n EMTP-RV envronment, whereas MATLAB scrpts have been used both to smulate the GPS staton of each slave and to develop the locaton procedure. The algorthm s appled to the solutons provded by EMTP,.e. set of voltage waveforms n the montored ponts of the network under test along wth the logc sgnals at the output of the relevant EDB. The analog devces of the slaves mplemented n EMTP envronment are reported n Fgure 4-7: they allow a relable performance analyss of the measurement system, n the sense that the equvalent crcuts of the devces n the measurement chan model satsfactory the actual behavour of the blocks. The nput sgnal v MV s condtoned by the block loss wth nomnal transformaton rato = 10,000 : 1, as the capactve voltage dvder nstalled n the measurement unt. The output sgnal of ths block drves a controlled voltage source (cv1) to obtan the output of the transducer (v n ). As the EDB s concerned, frst of all two DC components v off+ and v off-, equal magntude but opposte sgn, are superposed to the v n, then the two obtaned sgnals v n+ and v n- are sent to two comparators. The sgnal v f s the output of the low-pass flter R 1 /C 1 whose characterstcs are clarfed n the next chapter. The sgnal v f s compared to the two full-bandwdth sgnals v n+ and v n- by the two blocks compare whch emulate the operatonal amplfers used n the real EDB scheme confgured as comparators. The logc sgnals at the output of the couple of comparators are condtoned n order to obtan the two TTL-compatble output sgnals v out+ and v out-. The fnal output v out s a TTL sgnal whch changes ts level (0 5 V) every tme that one of the two v n waveforms cross the v f waveform. The voltage source v pont placed n seres to the controlled generator cv1 s used to superpose to the EDB nput the whte nose component. The sgnal v out comng from the EDB s mported n Matlab, where a functon s wrtten to smulate the GPS staton. As a matter of fact, the GPS scrpt reads a matrx reportng the EDB output n terms of two columns, one for the ampltude of each sample, the other one

121 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT for the correspondng samplng tme; a change of the status level corresponds to the frst non-zero value n the frst column, thus the relevant tme reference s regstered by the program. Fgure 4-7 Equvalent model of the EDB realzed n EMTP envronment The analyss of the propagaton of a transent dsturbance caused by a fault starts from steady state condtons. Even for ths network the EMTP smulaton has been confgured to determne the ampltude of the 50 Hz components of voltages and currents, and then

122 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT to take these data as ntal condtons n the tme-doman soluton. The ntegraton tme has been set equal to 100 ns - nomnal resoluton of the dstrbuted measurement system. Fgure 4-8 s a map of the overhead dstrbuton network under test, whch features a typcal radal topology consstng n a long man feeder (portons L1, L2, L3, L4) and n laterals connected wth t gettng to the load busses (lnes L5, L6, L7, L8, L9). The threephase power dstrbuton network s fed by a 150/20 kv staton; the source G represents the Hgh Voltage lne at ts prmary sde and the transformer T nomnal power and connecton type are 10 MVA and Y g /d respectvely. Each load bus conssts of a three phase balanced R-L mpedance, connected to the relevant feeder through a 20/0.4 kv transformer D/y g connected. As found n some lterature, e.g. [4], relevant to fault locaton technques, paraste capactors must be consdered n parallel to both sdes of each MV/lv transformer smulatng the behavour of transformers when sgnals havng frequency components of some hundreds of khz are concerned. Durng smulatons the power factor of all the loads was constant and assumed equal to 0.9, under the hypothess that all the load busses were provded wth an automatc capactor bank compensatng the phase shft between current and voltage at that pont. G Slave unt T S 7 Load Bus 1 L6: 5 km S 1 L1: 1 km S 2 L2: 3 km S 3 L7: 4 km S 8 Load Bus 2 L3: 5 km S 9 Load Bus 3 L8: 3 km S 4 L4: 4 km S 5 L5: 2 km L9: 1 km S 10 Load Bus 4 S 6 Load Bus 5 Fgure 4-8 Dstrbuton network under test (not n scale)

123 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT The EMTP Constant Parameter lne model has been adopted for the network lnes even n ths case. For the frst smulatons contnuously transposed three-phase lnes have been consdered, snce the tme doman soluton of the transent propagaton s easer under these condtons. Indeed, the symmetry between the conductors leads to only two equvalent propagaton modes nstead of three. Then, the smulatons have been repeated wthout ths confguraton, snce the Italan Medum Voltage system lnes are not contnuously transposed. The results obtaned n both cases are smlar, as shown n the followng. Fgure 4-9 shows the three phase-to-ground voltage waveforms acqured by the slave S1 at the occurrence of a short crcut between phases a and b 2500 m far from the measurement pont. The fault starts n correspondence of a lne-to-lne voltage v ab of 26 kv (the maxmum value) and ts equvalent resstance s zero n the frst case, equal to 9.5 kω n the second case. In the thrd graph the three phase voltages acqured by the same S1 slave are represented when the lne-to-lne fault has equvalent fault resstance 400 Ω and occurs when the v ab s 1.66 kv. How dfferent s the magntude of the transent dstorton n the three smulatons can be apprecated. It should be noted that the consdered faults feature extreme condtons as far as R f and V f are concerned. Before gong on wth the report of results, t can be useful makng clear the crtera adopted for the analyss of the performance of the proposed method by varyng the above parameters and condtons. In every smulaton campagn a gven fault has been smulated on each lne porton of the network under test. The confguraton of the network s fxed, whereas the boundary levels of fault parameters are looked for, so that nformaton on the lmts of the system are avalable. To do ths, the poston of the fault and the number of conductors nvolved s fxed; for the frst smulaton the heavest short crcut s consdered, hence zero fault resstance and occurrence of the fault n correspondence of the maxmum value of the voltage between faulted conductors. The locaton obtaned n such condtons s assumed as reference for the followng smulatons. Then, by keepng constant all the other parameters, the fault resstance s ncreased and the fault locaton procedure s appled. The result obtaned at the end of each smulaton s consdered correct only f the poston s the same one gven by the frst case. Amng at determnng the boundary value of the fault resstance that allows a correct locaton, R f s vared accordng to the bsecton algorthm: the boundary value (R f ) max s obtaned at the end of teratons wth dfferent R f values and corresponds to the condton: (R f ) max leads to correct result, (R f ) max + 100Ω leads to wrong result. Once the (R f ) max s found, the same smulatons sequence s repeated for a lower value of voltage between faulted conductors. For the parameter V f such a crtera for the varaton leads to a mnmum value,.e. that value that gve rse to a too low transent even f the R f s null

124 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT 4 x Va Vb Vc 2 1 V[V] t[s] x x 104 Va Vb Vc V[V] t[s] x x 104 Va Vb Vc V[V] t[s] x 10-3 Fgure 4-9 patterns of the transent dsturbances affectng lne voltages for dfferent lne-to-lne fault condtons

125 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Smulatons results Frst of all, smulatons have been run n whch the short-crcut pont was moved step by step along the faulted lne (L1) from nearby one end to the mdst, n order to determne the worst condton for the locaton n terms of dstance from fault pont to slave unt. In Fgure 4-10 the relevant results are reported by usng a characterstc R f /V f for the dfferent postons of the fault along L1. When the fault occurs n the mdst of the lne t s correctly detected and located wth the greatest dffculty. In order to reduce the number of smulatons to be run for the performance analyss of the method, the worst condton s assumed for the dfferent parameters, and thus durng smulatons the fault has been postoned always at the mdst of each one of the nne lnes. In Fgure 4-11 the results obtaned for a lne-to-ground fault on lne L6 by mplementng the prevous verson and the updated verson of the locaton procedure are reported n terms of the same R f /V f chart. The usefulness of the mprovements made n the latter one wth respect to the algorthm developed n Secton 3.4 s confrmed, n the sense that a much larger range of fault condtons can be correctly detected and located (red area plus blue area nstead of blue area). R f /V f for d = 500m R f /V f for d = 1500m R f /V f for d = 1000 m R f /V f for d = 2000 m R f /V f for d = 2500m R f [kω] 4 3,5 3 2,5 2 1,5 1 0, V f [kv] Fgure 4-10 Lmt fault condtons allowng correct results for dfferent values of the dstance faultslave

126 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Frst fault locaton algorthm updated fault locaton algorthm R f [Ω] f ,00E+00 2,00E+03 4,00E+03 6,00E+03 8,00E+03 1,00E+04 1,20E+04 1,40E+04 1,60E+04 1,80E+04 V f [V] Fgure 4-11 Comparson of fault condtons leadng to correct results when both versons of the locaton procedure are appled The type of faults consdered n the tests are phase-to-ground and phase-to-phase shortcrcuts and lghtnng strokes. They have been mplemented at the mdst of each lne of the network under test. The types of faults chosen for the smulatons have not to be consdered as a lmt for the valdty of the fnal performance of the fault locaton method. In fact other fault cases than the consdered ones would affect a hgher number of conductors (three-phase short crcut, two phase-to-ground short crcut, etc.) and hence would be heaver for the power system. In ths sense the chosen fault confguratons are representatve of the worst condtons for the dstrbuted detecton system, and the relevant performance lmts of the proposed method can be consdered relable for every other fault occurrence Drect lghtnng Fgure 4-12 reports the crcut used to model the lghtnng event n EMTP and the relevant current parameters. As far as lghtnng strokes are concerned, the nduced transent voltage always has so great ampltude that the measurement system detects the begnnng of the transent n all the slaves and the method gves correct results n all cases. Actually, n the smulated MV network there are no surge arresters nstalled n the nodes, and ther effect on the path of the propagatng transent s not taken nto account. The presence of such devces, anyway, cannot turn nto sgnfcant degradaton of the method operaton, snce surge arresters at the ends of the lne stroke by the drect lghtnng are desgned to start workng when the nduced overvoltage reaches values 3-4 tmes the nomnal voltage. For example, n Fgure 4-13 [5] the characterstc of a surge

127 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT arrester wth 8.5 kv ratng (.e. the lne-to-ground voltage level for a three phase system wth 15 kv nomnal lne-to-lne voltage) s reported. The surge arrester starts conductng the lghtnng nduced current when the relevant voltage gets to 30 kv. In any case the waveforms at the montored ponts of the network wll present a frst fast rsng front gettng to the peak value for sure much hgher than the EDB threshold. Fgure 4-12 Model of the Lghtnng fault Fgure 4-13 voltage-current characterstc of a surge arrester Phase to ground short crcut - Transposed lnes The maxmum value of fault resstance (R f ) for dfferent magntudes of the voltage u n the case of phase-to-ground faults s reported n Fgure Each symbol n the plot refers to a set of results relevant to faults located at the mdst of the same lne. In partcular, Fgure 4-14 shows the performance of the proposed method when the lnes are contnuously transposed

128 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT R f [Ω] L1 L2 L3 L4 L5 L6 L7 L8 L9 best ft lne L6 best ft lne L3 9.0E E E E E E E E E E E E E E E E E E E E+04 u' [V] Fgure 4.14 Grounded fault on transposed lnes: maxmum value of R f for dfferent u values lmt fault condtons allowng correct results for faults postoned at the mdst of each lne A man dfference can be seen between the results obtaned when the fault affects a lne connected to a transformer (the laterals feedng the load busses and the porton connected to the HV/MV staton) or a secton connected only to other lnes. When the fault affects the latter type of lnes, the method performance s better n the sense that hgher R f values let the procedure compute a correct locaton. The presence of a transformer at one end of a lne secton seems to have a strong nfluence on the boundary fault condtons, ndeed the ponts n the graph representng results obtaned when the faulted lne s a lateral are almost superposed. On the contrary the ponts relevant to the cases n whch the fault s located along the feeder are not concdent. The nfluence of the length of the faulted secton can be apprecated: the method performs better when the fault affects lne L3 (5 km), then lne L4 (4 km) and fnally L2 (2 km). Moreover, the effcency of the proposed detecton and locaton method has been checked by comparng the maxma values of R f n Fgure 4-14 to actual equvalent resstance values for the same type of fault. In 2000 the man Italan electrcty utlty carred out a measurement campagn on the values of fault resstance, whose report s represented n Fgure It can be easly seen that over 50% of grounded short crcuts featured an R f value lower than 10 Ω; about 90% of them had R f <1 kω. The smulaton results have been compared to the cumulated frequency of real values of R f for the same type of fault. In order to defne the characterstcs of a new automatc protecton equpment to be nstalled n MV networks, n partcular dealng wth a threshold settng n case of grounded short crcuts, the man Italan utlty carred out a measurement campagn. In 2000, measurements amng at obtanng the frequency

129 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT dstrbuton of R f values were made by analyzng the data regstered by the protecton equpments at the tme of nterventon. The results of such measurements (around 1000 acqustons) are reported n Fgure 4-15 n terms of frequency dstrbuton and cumulated frequency as well. Table 4-5 Fault parameters relevant to the ponts reported n Fgure 4-14 L2 L3 L4 L6 u [V] R g [Ω] R g [Ω] R g [Ω] R g [Ω] 1,54E+04 2,85E+03 1,49E+04 7,40E+03 1,47E+04 7,40E+03 1,52E+04 8,40E+03 8,77E+03 1,70E+03 4,44E ,66E+03 5,40E+03 4,34E+03 2,70E+03 8,49E+03 4,70E+03 4,28E+03 2,30E+03 8,41E+03 4,70E+03 4,28E+03 2,40E+03 frequency Cumulated frequency Resstance [Ω] of the phase to ground fault Fgure 4-15 measurement results of the Italan utlty campagn on grounded faults It should be notced that the values reported n the hstogram are not the actual fault resstances, but the equvalent ones seen by the protecton equpment, whose value can depend on the protectve settngs of each devce. Fgure 4-16 reports the percentage of phase-to-ground faults regstered by the utlty that could have been correctly detected and located by the proposed method. Each colour of

130 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT the area relevant to fault condtons leadng to a correct locaton corresponds to a partcular value of cumulated frequency of R f n Fgure The comparson shows the hgh effcency of the procedure n locatng faults. 9.0E+03 R f [Ω] 8.0E+03 L1 L2 L3 L4 L5 L6 L7 L8 L9 best ft lne L6 best ft lne L3 7.0E E E E E E E E E E E E E E E E E E+04 Cumulated frequency of R f values of phase-to-ground faults obtaned by a measurement campagn on electrc power dstrbuton systems; comparson wth the best performance of the proposed method n terms of percentage of detected faults as R f ncreases: u' [V] >98%; 98%; >91%; >85%. Fgure 4-16 Comparson of real values of grouded-fault resstance regstered n MV systems wth the range of R f values that allow the method to correctly locate the fault Phase-to-ground short crcut - Untransposed lnes Table 4-7 Fault parameters relevant to the ponts represented n Fgure 4-17 L2 L3 L4 L6 u [V] R f [Ω] R f [Ω] R f [Ω] R f [Ω] 1,54E+04 1,49E+04 7,40E+03 1,47E+04 7,40E+03 1,52E+04 8,40E+03 8,77E+03 2,85E+03 1,70E+03 4,44E ,66E+03 5,40E+03 4,34E+03 2,70E+03 8,49E+03 4,70E+03 4,28E+03 2,30E+03 8,41E+03 4,70E+03 4,28E+03 2,40E

131 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT L1 L2 L3 L4 L5 L6 L7 L8 L9 9.0E E E E+03 Rf [Ohm] 5.0E E E E E E E E E E E E E E E E+04 Fgure 4-17 lne-to-ground fault on untransposed lnes: maxmum value of R f for dfferent u values. Lmt condtons allowng correct results for faults postoned at the mdst of each lne. u' [V] 1.2E+04 L1 L2 L3 L4 L5 L6 L7 L8 L9 best ft lne L1 best ft lne L6 1.0E+04 y = 0.38x E+03 Rf [Ohm] 6.0E+03 y = 0.29x E E E E E E E E E E+04 u 'LL [V] Fgure 4-18 lne-to-lne fault on untransposed lnes: maxmum value of R f for dfferent u LL values lmt fault condtons allowng correct results for faults postoned at the mdst of each lne Fgure 4-17 shows the method performance for phase-to-ground faults n the case of untransposed overhead lnes n the network. The method performs well also n ths confguraton of the conductors, as t can be seen by comparng Fgures 4-17 and

132 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT The results relevant to both cases agree, except for the opposte dependence of the lmt value of R f on the lnes length Phase-to-phase short crcut - Untransposed lnes Fgure 4-18 reports the maxmum value of R f n functon of the lne-to-lne voltage nstead of the phase voltage gven that phase-to-phase short crcuts are consdered. For ths type of fault no remarkable dfference can be found n the results when the man feeder or one lateral lne s faulted Varatons n the network topology Dsconnecton of lnes Smulatons have then been run under dfferent operatng condtons. Frst of all the performance of the method have been tested for phase-to-ground faults wth dsconnected lnes respect to the confguraton of the network regstered n the master database. The system confguratons taken nto account for smulatons are the followng: Fault on lne L1 when the lne L6 s not connected; Fault on lne L2 when the lne L6 s not connected; Fault on lne L2 when the lne L7 s not connected; Fault on lne L3 when the lne L6 s not connected; Fault on lne L3 when the lne L8 s not connected; Fault on lne L4 when the lne L8 s not connected; Fault on lne L5 when the lne L9 s not connected; Fault on lne L4 when the lne L9 s not connected; Fault on lne L6 when the lne L7 s not connected; Fault on lne L8 when the lne L7 s not connected. By analyzng the obtaned results dfferent behavour of the method can be drawn: ) f the faulted lne s not connected to a node where also the dsconnected lateral ends, no dfference n the results can be apprecated; ) f the fault affects a lne that feeds a load bus, and hence connected to a transformer n ts end, the maxmum value of R f leadng to a correct locaton s almost dentcal; ) f the fault occurs on the man feeder the boundary value of R f s lowered of around 40% respect to the value relevant to the same fault case n normal operatng condtons. Consderatons ) and ) are vald ndependently of whch lne havng one common end wth the faulted one s dsconnected

133 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Table 4-8 Fault parameters relevant to the ponts represented n Fgure 4.18 L1 L2 L3 L4 L5 L6 u LL [V] R f [Ω] R f [Ω] R f [Ω] R f [Ω] R f [Ω] R f [Ω] 1,54E+04 2,85E+03 1,55E+04 3,00E+03 1,49E+04 7,40E+03 1,47E+04 7,40E+03 1,47E+04 3,00E+03 1,52E+04 8,40E+03 8,77E+03 1,70E+03 4,44E ,84E+03 1,90E+03 4,48E ,66E+03 5,40E+03 4,34E+03 2,70E+03 8,49E+03 4,70E+03 4,28E+03 2,30E+03 8,41E+03 4,70E+03 4,28E+03 2,40E+03 8,52E+03 1,90E+03 4,41E Dsconnecton of load busses Some smulatons of phase to ground fault n the case of loads not connected to ther busses have been run. The performance of the proposed fault locaton method are dentcal to the ones shown above, both f the fault occurs on the lateral wthout load at ts end and f the fault affects a lne connected to the unloaded lateral. Varatons of loads power Smulatons results have shown that decreasng the values of resstance and nductance of any three phase load tll 50% can lead to varatons of the boundary value of R f relevant to a grounded fault of some Ohms (whereas R f s n the order of some kloohms). The same can be sad f the power of the load s ncreased of 50%. In order to obtan apprecable varatons n the performance of the fault locaton method (anyway R f values decreased only of some ponts %) the load power has to be vared of one order of magntude. Fnally, the smulatons results lead to conclude that the method s not nfluenced by varatons of the power absorbed by the loads

134 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT Non lnear loads harmonc dstorton The effect of non-lnear loads has also been consdered. Odd harmonc frequency components up to 11 th order have been smulated n the network and the boundary magntude levels descrbed n [6] have been assumed for each order. As a matter of fact, n the IEC standard the hghest tolerated level of harmonc current polluton corresponds for every frequency component to the value I 1 /h, where I 1 s the nomnal RMS of the fundamental component of the current and h s the harmonc order of the frequency component. 2 x V[V] t[s] Fgure 4-19 voltage waveforms when non lnear loads are present n the network. Ideal current sources have been connected n parallel to each load bus to generate harmoncs to check whether the slaves could always dstngush the transent dsturbance generated by a fault from harmonc dstorton (perodc dsturbance). For example, Fgure 4-19 reports the voltage waveforms acqured by the slave unt S2 when non lnear loads are present n the electrc system. As expected, no meanngful dfference n the performance of the detecton and locaton system can be found, n fact the EDB nstalled n each slave unt s mmune to any frequency component of the voltage sgnal lower than 10 khz, and hence no fake event detecton can happen n the presence of perodc dstorton. At the same tme, the transent dsturbance s always dstngushed by the harmoncs superposed to the 50 Hz phase voltage and mmedately regstered by the remote statons

135 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT 4.3. Implementaton of a radal dstrbuton network to test the verson of the fault locaton procedure ntegrated wth DWT analyss Smulatons have been run n EMTP-RV [3] envronment to obtan the voltage waveform on a dstrbuton network when the electromagnetc transent due to a fault propagates through the lnes. The sgnals acqured n dfferent ponts of the network under test have been mported n MatLab envronment, where proper scrpts emulate all the nstrumentaton requred by the dstrbuted system. These scrpts process the set of dgtal waveforms relevant to one case of fault smulaton (as f they were the data acqured by the DAQ board and sent from slave unts) and mplement the fault detecton and locaton algorthm. Fault locaton Load Bus 2 Remote staton L4: 2 km G T d 1 d 2 S 2 S L5: 5 km S 4 3 S 1 L1: 2 km L2: 3 km d 5 Load Bus 1 L3: 1 km d 3 d 4 Load Bus 3 Fgure 4-20 The power dstrbuton network used for tests (not n scale) Fgure 4-20 s a map of the power dstrbuton system mplemented n EMTP smulatons. It conssts n a 10-km long man feeder (lnes L1, L2 and L5) and n two laterals: L4 (2-km long) and L3 (1-km long). In the system three-phase overhead lnes are used. The model adopted for the lnes s the Constant Parameters Lne Model,.e. an equvalent crcut wth unformly dstrbuted parameters R, L, C and G per unt length. The radal dstrbuton network s fed by a 150/20 kv substaton G, where the transformer T s of the Y g /d connecton type. All the load busses are modeled as a three-phase balanced mpedance. Each load s connected to the relevant lateral through a 20/0.4 kv transformer. As suggested n lterature [4] capactors are connected n parallel to each wrng of the EMTP transformer model; ths way a frst order approxmaton of ts behavor when sgnals nclude frequency components around 100 khz. Three-phase and sngle phase-to-ground short crcuts have been smulated at dfferent locatons, shown by flashes n Fgure 4-20, by usng deal swtches. Four slave unts of the dstrbuted measurement system, denoted by S 1, S 2, S 3 and S 4, are placed n correspondence of the black bullets n Fgure

136 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT The ntegraton tme for all the faulted network smulatons has been set to 100 ns, equvalent to the nomnal resoluton n tme performed by the measurement chan of a slave unt. The values of T nkref relevant to the propagaton along the man feeder have been obtaned by dvdng the nomnal length of each secton by the propagaton speed of the α mode. The proposed algorthm detects correctly the lne affected by the fault for the postons consdered and grounded faults are dstngushed from ungrounded ones. The man feeder s assumed to be faulted n the case of lnes L 1, L 2, L 5 affected by short crcuts, whereas not faulted f the event occurs on lnes L 3 and L 4. As a consequence, the locaton of the fault s estmated by dfferent algorthms for the two set of postons, and the relevant results are shown n Table 4-9. Table 4-9 Estmated and actual dstance of the fault from the nearest measurement unt Knd of fault Three phase Phase to ground d 1 (m) d 2 (m) d 3 (m) d 4 (m) d 5 (m) Estm. Ref. Estm. Ref. Estm. Ref. Estm. Ref. Estm. Ref As far as symmetrc short crcuts are concerned, t can be stated that the fault s located wth very good accuracy by both the procedures, whereas n case of phase-to-ground faults the analyss based on the travelng waves theory features worse accuracy n locatng the short crcut along the faulted lateral. Ths s due to the dffculty n detectng n the transent waveform the correct spke used to compute the dstance between fault pont and lne head, correspondng to the wave reflected from the fault pont that has to be dstngushed by all the other smultaneous reflectons comng from the nodes of the network. moreover, losses along the lne dump qute quckly the magntude of the oscllatons, hence f the fault s not very heavy for the network t does not gve rse to a dsturbance much hgher than the usual nose level affectng phase sgnals. Analyss of the DWT decomposed sgnals under dfferent ponts of vew have been performed nvestgatng on the chance to overcome the problem. For example other levels of detals have been analyzed, the energy contrbuton of the levels have been compared, snce each decomposton level s relevant to a frequency subband and the equvalent frequency of the oscllatons n the transent depends on the dstance fault-node. The sgnals have also been reported n the tme doman by applyng the nverse wavelet transform only to the hghest levels so that only the waveshape of the dsturbance s obtaned. Fnally, the DWT has been appled to the modal voltages nstead of the phase ones, snce grounded short crcuts as dssymmetrc condtons lead to a

137 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT hgher zero mode couplng respect to the three phase ones. All such approaches do not lead to a sgnfcant mprovement of performance. level central freq.3571khz- Vpk= level central freq.1786khz- Vpk= level central freq.893khz- Vpk= level central freq.446khz- Vpk= level5 x 10 4 central freq.223khz- Vpk= level x 10 4 central freq.112khz- Vpk= level7 x central freq.56khz- Vpk= level8 x central freq.28khz- Vpk= x 10 4 central freq.14khz- Vpk= x 10 4 central freq.7khz- Vpk= level level level11 x 10 4 central freq.3khz- Vpk= level12 x central freq.2khz- Vpk= Fgure 4-21 Phase-to-ground fault n the lne L 4 : detals coeffcents obtaned by processng the voltage sgnal measured n S

138 4. ANALYSIS OF THE PERFORMANCE FEATURED BY THE FAULT LOCATION METHOD SIMULATIONS OF DISTRIBUTION NETWORKS IN EMTP ENVIRONMENT References [1] IEEE Dstrbuton plannng WG report, Radal dstrbuton test feeders, IEEE Trans. on Power Systems, vol.6, n.3, pp , August 1991; [2] IEC Internatonal Electrotechncal Vocabulary Electromagnetc Compatblty, Internatonal Electrotechncal Commsson, Geneva, Swtzerland, 1997; [3] J. Mahseredjan, L. Dubé, L. Gérn-Lajoe, New advances n the Smulaton of Transents wth EMTP: Computaton and Vsualzaton Technques, Electrmacs, 19 August 2002; [4] A. Borghett, S. Cors, C.A. Nucc, M. Paolone, L. Peretto, R. Tnarell, On the use of contnuous-wavelet transform for fault locaton n dstrbuton power networks, Internatonal Journal of Electrcal Power & Energy Systems, vol. 28, n. 9, November 2006, pp ; [5] Nakada K.; Yokota T.; Yokoyama S.; Asakawa A.; Nakamura M.; Tanguch H.; Hashmoto A., Energy Absorpton of Surge Arresters on Power Dstrbuton Lnes due to drect lghtnng strokes effects of an overhead ground wre and nstallaton poston of surge arresters, IEEE Trans. On Power Delvery, Vol. 12, No. 4, October 1997, pages ; [6] EN Electromagnetc Compatblty (EMC) Part 2-4: Envronment Compatblty levels n ndustral plants for low-frequency conducted dsturbances, CENELEC, Bruxelles, Belgum,

139 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5. The dstrbuted measurement system for transents detecton The measurement technques of the man ndexes of PQ have been the subject of specfc standard defntons n order to let expermental results of dfferent measurement campagns be comparable. Thanks to the new hardware and software technologes, the nstrumentaton s more and more effcent, as well as less easy for the operator to use and to nterpret the result goodness. A sngle nstrument can be set to measure dfferent type of dsturbance just by changng the elaboraton algorthm of the nput data. The ssue s not fndng the rght nstrument, usually avalable for each applcaton, but the best devce for condtonng the nput sgnals. In fact transducers have to transfer voltage and/or current to the measurement nstrumentaton wthout modfyng the man characterstcs of ther frequency spectrum. Usual commercal nstrumentaton s mult-purpose, for example dsplay n real tme the frequency spectrum of the nput under the hypothess of perodc sgnals; gve statstcal parameters regardng transent dsturbances; can log data for lmted tme ntervals. The most sgnfcant bandwdth of the sgnal to be transduced can be n the order of some klohertz as well as hundred of klohertz accordng to the phenomena to be montored. The dynamc range of the nput sgnals can be order of magntude greater than the electronc equpment amed at elaboratng them, hence the adequate solaton level has to be granted by the transducer nterposed between the two blocks. The tradtonal TA and TV (.e. current and voltage transformers) feature a too low frequency bandwdth. In ths context nowadays electronc devces, such as Hall-effect probe and Rogowsk col, are substtutng passve transducers. When Dgtal Sgnal Processng (DSP) s nvolved, a basc aspect s the choce of the samplng technque accordng to the measurement condtons. The samplng frequency has to be chosen n a range that avods alasng error,.e. the brth of fake low frequency components due to a too low samplng frequency respect to the nput spectrum. The tme nterval for the acquston of the sgnal has to be extended n functon of the desred frequency resoluton: hgh resoluton n the sgnal spectrum requres a long tme wndow, but such a condton s acheved only n case of stable sgnals. The choce of the measurement strategy s not so easy and experence s needed, even because commercal nstrumentaton s characterzed by the manufacturer wthout declarng all the operatng condtons. The measurement system desgned to montor the voltage qualty of a dstrbuton power system and to mplement the method for the locaton of the transent source descrbed n

140 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION chapter 3 s a dstrbuted measurement system. The archtecture chosen for the applcaton s master slave, where a set of slave statons are nstalled n the network nodes and are amed at regsterng data at the occurrence of a transent event and send them to the master unt va GPRS. In practce the communcaton net let each slave staton send data n real-tme to the master unt n case of anomalous workng condtons of the power system, but also the master unt send nformaton to any slave for dagnoss purposes or check up of the hardware. Every slave unt s a stand-alone devce, n the sense that a mcroprocessor s programmed to control the system settngs, the acquston of data and the communcaton wth the master. Let s recall the block dagram of a remote unt (slave), reported n Fgure 5-1. It both acqures the startng nstant and the waveform of a transent. The former nformaton s then sent to the master unt whch locates the source of the transent by processng the data receved from all the slaves. The dashed box n Fgure 5-1 contans the blocks performng the measurement of the transent startng nstant. Only by way of example let us consder the problem of locatng the source of a transent voltage. The voltage u(t) at the montored pont s condtoned by a Voltage-to- Voltage Transducer (VVT), whose output u VVT feeds an Event Detecton Block (EDB). The EDB output u out s a logc sgnal that, as a transent occurs, trggers both a Data AcQuston board (DAQ) and a GPS-based devce, whch provdes the relevant tme stamp. The tme stamp s determned wth an uncertanty that propagates through the algorthm whch the proposed measurement procedure reles on, and affects the estmate of the transent source locaton. Input voltage u VVT u VVT EDB u out GPS-based devce PC tmestamp Transent pattern Trgger n DAQ Fgure 5-1. Block dagram of a remote unt (slave). In a qute smlar way the measurement system can also locate the source of transent currents. In such a case the output of a sutable current-to-voltage transducer n place of u VVT has to be consdered. In the followng the blocks of one measurement channel are consdered one by one and descrbed

141 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5.1. The voltage transducer The voltage to voltage transducer chosen for the system s a capactve voltage dvder made by Pearson Electroncs, the model VD305A (see Fgure 5-2). The reason why such a devce has been chosen for power qualty measurement n dstrbuton networks are: large bandwdth, hgh accuracy and lnear nput-output characterstc for a wde range of nput voltages, low delay on the output sgnal respect to the nput one, mmunty to electromagnetc dsturbances. Such a performance let the transent dsturbance affectng the hgh voltage sgnal be correctly transduced on the low voltage sde, feedng both the Event Detecton Block and the DAQ board. It s not hard to understand that the characterstcs of the condtonng block are of prmary mportance n the desgn of the measurement chan. The Model VD-305A capactve voltage dvder s ntended for the measurement of voltage ampltude and wave-shape of ac sgnals at hgh potental [1]. It has a nomnal dvson rato of 3800:1, and the exact measured rato s prnted on the name-plate. Ths rato s measured n nsulatng ol at 35 C, and s accurate to ±5%. The dvson rato s temperature compensated to ±1% over the range of 20 to 80 C. The unt conssts of two capactors connected n seres. The hgh voltage center electrode forms a capactor wth a guarded pckup rng located n the lower metal cylnder. Ths pckup rng s connected to the center conductor of the output connector va a 50 Ohm resstor. The low-voltage capactor connects the pckup rng to the outer conductor of the connector. The output voltage s thus a fracton of the nput voltage determned by the rato of the capactances. The standard calbraton s for use n ol, whch has a delectrc constant of about 2.3. If the unt s used n ar, ts dvson rato wll be approxmately 11500:1, but f ths s the ntended use, a factory calbraton of 5000:1 n ar would be more accurate and desrable. The maxmum pulse voltage ratng for use n ar s 50 kv. These ratngs are for pulses up to 5 μs duraton. Fgure 5-2. The capactve voltage dvder

142 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5.2. The event detecton block The Event Detecton Block s an analog crcut based upon two comparators. In order to detect n real tme any transent dstorton affectng a sgnal, both mpulsve or oscllatory, the nput u n s compared to the sgnal obtaned by flterng u n tself and by superposng to the fltered sgnal u f a bas. The block dagram of the EDB s shown n Fgure 5-3. the offset component summed to u f s postve n one case and negatve n the other case, leadng to u + - f and u f respectvely. In practce the pattern of the orgnal sgnal u n s nstantaneously ncluded wthn a couple of sgnals that represent a double threshold for ts hgh frequency content. The sgnals at the nput and output of the two comparators are represented n Fgure 5-4. Comparator #1 changes the level of ts output u + out when a surge startng wth a rsng front affects the nput voltage, whereas comparator #2 turns ts level down when the transent presents a fallng front. An And-gate s used to make the logc sgnal at the EDB output change ts level for any commutaton of u + out and u - out,.e. n correspondence of the frst peak of the dsturbance, ndependently of ts dervatve sgn. Snce the transents are most of the tmes of the oscllatory type, t s easy to understand that such an operatng prncple would lead to pulse sequences on u out because t would commute n correspondence of every oscllaton of the dsturbance. By consderng the entre measurement chan of a channel, ths behavour would correspond to gve a repettve trgger to the DAQ board and the GPS staton, as f more than one event occurred at once. Moreover, the devces are busy for a tme nterval much longer respect to the perod of oscllatons, tme needed to acqure, convert and save the waveform and the tmestamp, but also the PC takes tme to save the data n ts memory support, compress and send them to the master va GPRS. +V cc Resstve voltage dvder 1 Sum 1 u f + Comp. 1 u out + u n -V cc Low pass flter Resstve voltage dvder 2 u f Sum 2 u f - Comp. 2 u out - & gate Mono stable u out Fgure 5-3. Block dagram of the analog crcut of the EDB

143 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION u f + u f + u n u f - u n u f - u out - u out + u out - u out + u out u out Fgure 5-4. Sgnals on the EDB crcut at the occurrence of a postve or negatve surge on u n. On the bass of these consderatons, the EDB s blnded for a tme nterval correspondng to 1,5 s after each detecton of an event. To ths purpose, a monostable multvbrator s connected at the output of the And gate. Fnally, u out commutes to low level synchronously wth the frst commutaton of one of the two comparators and keeps that value for 1,5 s. In ths way the electronc equpment s trggered once by u out and can carry out the acquston process. A basc step n the desgn of the crcut s the choce of the cut-off frequency of the lowpass flter. Ths frequency has to be much greater than the maxmum harmonc component that can affect the power system voltage, but at the same tme lower than the typcal spectrum assocated to transent dsturbances. In ths way the logc sgnal u out trggers the GPS and DAQ board only n case of non perodc dsturbances, and detects any type of transent. Under such operatng condtons the cut-off frequency should be just a few greater than 2 khz,.e. the value correspondng to the 40 th harmonc order of the 50 Hz component, so that even the slowest transents are taken nto account by the slave. By consderng the sgnals u + - f and u f compared to u n, the postve and negatve bas superposed to each of them to shft sgnals along the vertcal axs can be seen as the real threshold of the EDB. Settng the magntude of such bases turns nto settng the senstvty of the detector; the deal confguraton of the crcut would be makng the EDB more sensble as possble, decreasng the bas tll a value near zero; n real applcatons t s not possble to reach such performance of the transent detector, because the nput sgnal n steady state s not a perfectly clean snusodal waveform. Hence, n order to

144 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION avod fake detectons and acqustons of the measurement system, the threshold has to be greater that the maxmum peak that can be seen n the whte nose affectng the network voltages. By consderng the defnton of nose accordng to the standards and at the same tme voltage waveforms acqured durng smlar measurement campagns carred out on dstrbuton networks, the nose reveals to be n the order of 1% of the magntude of the fundamental component, and ts frequency content s lower than 300 khz. The bas n the EDB has been set equal to 2% of u n ampltude; t should be noted that the frequency spectrum of the nose covers all the frequency range n whch EM transents can be found, so fnally t can be sad that the presence of nose leads to worse performance of the measurement system by puttng a hgher boundary for ts senstvty. u VVT 100 k 3.3 k 1.2 k +V cc O A 1.2 k 160 n 1 M 1 k 1 k + - B 1 k u + f + - C 1.2 k & monost u out -V cc 100 k 3.3 k O + 63 n 1 k 1 k + - D 1 k u - f + - E 1.2 k Fgure 5-5. Scheme of the Event Detecton Block. The entre procedure above descrbed and mplemented n the analog crcut of the block works correctly and makes sense only f the sgnals u f compared to u n are always n phase wth t. Ths s a crucal aspect of the practcal realzaton of the devce, n fact the cut-off frequency adopted for the low-pass flter s not so hgh respect to the fundamental component of the nput, hence the output of the flter s affected by a not neglgble phase delay. As explaned n secton 3.5, the nput sgnal u n has to be compared to a sgnal reproducng only ts low frequency content, elmnatng as much as possble the transent dsturbance superposed to t. f ths condton s not respected, the detecton of the begnnng of the transent s affected by an error that leads to wrong results when the fault locaton algorthm s appled. Fnally, the cut-off frequency of the low-pass flter cannot be ncremented to reduce the phase-shft affectng the 50 Hz component, because the effcency of the flter tself would be strongly compromsed. To brng back to zero the delay between nput and output a dfferent soluton has been then adopted: a hgh-pass

145 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION flter n cascade wth the low-pass one has been added, havng ts same order. The cutoff frequency of the hgh-pass flter, much lower than 50 Hz, s chosen so that the phase of the global output u f s null referred to u n. The dodes nserted at the output of each comparator are used to lmt the dynamc of the logc sgnal. In fact the comparator commutes from +5 V to -5 V whereas the TTL compatble sgnal that has to be sent to the sequent devces as external trgger has to change level from +5 V to 0 V n negatve logc The GPS-based devce What s GPS? The frst satellte of the GPS was launched nto orbt n The am of the GPS was to replace the old, naccurate, and nconvenent system of satellte navgaton called transt, put nto operaton n 1959 to control Polars rockets. In March 1994 the U.S. Department of Defense announced a prelmnary permsson to use the GPS and t was declared fully operatonal on July 1995 [2]. The GPS constellaton conssts of 24 solar-powered satelltes that orbt the earth n 12 hours. They are equally spaced on sx crcular orbts, 60 apart, about 20,183 km above the earth s surface. The orbts are nclned at about 55 respect to the equatoral plane. Ths constellaton allows concurrent communcaton wth at least four satelltes at any one-tme, enablng to assess the coordnates (lattude and longtude for sea vessels and land vehcles) or three coordnates (addtonally alttude for arcrafts). In ths way also the accuracy of the estmaton s allowed. In practce, GPS recevers located between 80 N and 80 S not occluded by clffs, mountans, skyscrapers, etc. can smultaneously receve sgnals from fve to nne satelltes. The sgnal transmtted by satelltes comprses ts dentfcaton code, a GS tme stamp and locaton nformaton, such as locaton of other satelltes. The recever analyzes the data and rejects records from satelltes that are less than 10 to 15 above the horzon and then uses the transmsson from the satelltes n the optmal geometrc confguraton to calculate the maxmum postonng accuracy. The recever then dsplays the result as geographcal coordnates. The geographcal poston calculaton s based on the precse measurement of a dstance between a GPS recever and the satelltes. The locatons of the satelltes are constantly tracked by ground statons and are known exactly at all tmes. The dstance measurement s based upon the accurate measurement of ntervals between sgnal transmsson from a satellte and ts recept by a recever, whch s nstalled, for example, on a shp or a car whose poston s beng computed. Each satellte of the system s

146 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION equpped wth four atomc clocks whch automatcally correct the quartz clock of the GPS recevers Untl May 2000, two GPS postonng servces of dfferent precson were avalable. The more accurate one, called Precse Postonng Servce, was based on an encrypted sgnal called P-Code transmtted at a carrer frequency of 1227,60 MHz. the poston estmaton error was not greater than 18 m horzontally and 23 m vertcally. PPS was avalable only for the U.S. mltary. The less precse servce, called Standard Postonng Servce (SPS), was avalable for all cvl users worldwde wthout charge or restrctons. A coarse acquston code was transmtted from a satellte at 1575,42 MHz. the accuracy of the SPS was ntentonally degraded by the U.S. Department of Defense. An abundance of error estmaton methods for satellte navgaton systems n general makes ther comparson and evaluaton dffcult. Some sources cte root-mean-squared error of consecutve measured postons; others rely on selectve measurements n a gven perod. Above all, ths error s often calculated wth a dfferent confdence level, t may refer to the real recever s poston (from a chart) or to the average poston of all the conducted measurements. The error values are gven as naccuracy n lattude and longtude or as a dstance between real and determned poston. Accordng to the U.S. Department of Defense, the SPS horzontal accuracy s ± 100 m at 99.5% confdence level and ± 300 m at 99.9%. the man component of GPS postonng naccuracy results from tme and space varyng condtons of rado wave propagaton, whch both depend on atmosphercs, dsturbances n the satelltes, orbt stablty, and so on. The error due to the transmtter and recever operaton precson or tme measurement accuracy s neglgble. A method for GPS precson enhancement commonly used by cvl users s the Dfferental Global Postonng System. It s based on addtonal GPS measurements of a precse poston from a nearby correctng staton. The error, whch s characterstc for a large regon around the staton, s rado transmtted and taken nto account by the DGPS recevers as a poston correcton The GPS staton The operaton of the devce reles on a programmable GPS 168 PCI card by Menberg. The staton s made of two devces: the GPS antenna (see Fgure 5-6) and a PCI board (see Fgure 5-7) lnked by a BNC cable. In the rear panel of the board, a part from the BNC connector, two led and a DB9 connector are found. The red led turns on every tme the board s started and turns off only at the end of the synchronzaton process carred out by the recever, whereas the green led turns on when the poston of the recever s computed

147 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Fgure 5-6. The antenna of the GPS staton. Fgure 5-7. The PCI board of the GPS system. The DB9 connector, whose pns are reported n Table 5-1, cannot be connected to a seral port of the PC because n ths case the voltage level s ±12 V, whereas the rado clock for example operates wth a TTL logc (0 5 V). the DB9 connector s useful to let nput and output of the staton be avalable for the operator. Table 5-1. GPS DB9 connector. Pn Sgnal Dl2 1 +5V 1 2 RxD - 3 TxD - 4 P_MIN MHz 10 5 GND - 6 CAP0 2 7 CAP1 3 8 P_SEC 4 9 DCF_OUT 6 By usng the correspondng DIL swtch, some nput (such as User Capture 0 and 1) and output (such as Pulse per Second/Mnute) of the PCI board can be actvated. The sgnals wthout a DIL swtch are always avalable. In the case of nterest, the two Tme Capture Input and the Pulse per Second output have been turned on. The frst ones are amed at detectng events; n dong ths the tme resoluton of the GPS s 100 ns. The event s detected accordng to the TTL logc b the User Capture n ts fallng front and then saved n a buffer. Once t has been saved, the event data can be sent va the Com Port of the GPS or read by the PCI bus n order to be reported n the montorng software. Menberg

148 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Rado Clock s the name of the program gven by the manufacturer for montorng and dagnostc of the GPS unt. For the applcaton n the dstrbuted measurement system, n order to have a complete lst of all the tmes when an event occurred, the relevant data have been saved n a text fle by usng Hyper termnal of Wndows. The data are transmtted through a GPS COM Port wth a FIFO strategy mmedately after the event occurrence, thus there s no rsk of saturaton for the memory buffer of the unt. It should be noted that the GPS staton cannot detect two consecutve events separated by less than 1.5 ms of delay; n ths case a warnng message s dsplayed n the software nterface: capture overrun. The Pulse per Second, reported n Fgure 5-8, s a sgnal wth 1s perod. The hgh level (5 V) last 200 ms, then the low level (0 V) last the remanng 800 ms; the sgnal features ts best accuracy, ± 250 us, after the synchronzaton and 20 mnutes of contnuatve workng. In the frst expermental phase for the development of the measurement chan ths logc sgnal played a man role as tme reference for the calbraton procedures. Hyper Termnal of Wndows s a software that lets data be acqured through the Seral Port of a PC, so a partcular connector has been realzed to lnk the GPS Com Port to the RS-232 Port of the PC. In order to let Hyper Termnal acqure data, a sesson confgured n the followng way must be started: ) data transmsson speed = bps; ) bt number per sngle data = 8; stop bt = 1; flux control = none. Coherent confguraton has to be set n the GPS Com Port. Fgure 5-8. The Pulse per Second sgnal acqured by the DSO. Menberg Rado Clock Montor s an applcaton that let the operator check the status of the reference clock,.e. the tme reference of the PC and the tme adjustment servce. The latter one gves nformaton about the correcton factor appled to the PC tme reference to synchronze t wth the rado Clock The DAQ Board The waveshape of each transent dsturbance affectng the voltage sgnal s saved nto a fle by an acquston and dgtal converson board. The model chosen for the purpose s

149 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION the ADC-212/100 by Pcotech (see Fgure 5-9). Two dfferent software can be nstalled on the PC n order to control and communcate wth the DAQ board: Pcoscope, whch enables the desktop PC to work as a Dgtal Storage Osclloscope hghend, a spectrum analyzer and a multmeter; PcoLog Player, whch allows to use the DAQ board as a devce to acqure data n a fast way. The man specfcatons of the devce are reported n Table 5-2. the DAQ board can be connected to the PC through the parallel port or by usng an adaptor Pco/USB port. Table 5-2. Pcotech ADC-212/100 data sheet. Channels 2 BNC + 1 external trgger Frequency bandwdth 50 MHz Samplng Frequency (sngle channel) 100 MS/s Samplng Frequency (dual channel) 50 MS/s Samplng Rate (repettve sgnals) 5 GS/s Resoluton 12 bt Buffer Sze 128 K Dynamc range 80 db Scope tmebase 100ns/dv to 50s/dv Spectrum range 0 to 50 MHz Trgger Modes Free run, Auto, Repeat, Sngle Pre/post trgger ±100% Voltage range ± 50mV to ±20V n nne steps Overload protecton ±100 V Input Impedance 1 MΩ Couplng AC, DC Accuracy ±1% Power supply V (man adaptor suppled) Output connector D25 to PC parallel port (cable suppled) Dmensons (140 x 190 x 45) mm The confguraton and control of the board can be carred out by usng the dedcated software provded by the manufacturer as well as programs developed n LabVew envronment

150 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Fgure 5-9. Acquston and Analog to Dgtal converson board The GSM protocol Global System for Moble communcatons (GSM: orgnally from Groupe Spécal Moble) s the most popular standard for moble phones n the world. Its promoter, the GSM Assocaton, estmates that 82% of the global moble market uses the standard. GSM s used by over 2 bllon people across more than 212 countres and terrtores. GSM dffers from ts predecessors n that both sgnallng and speech channels are dgtal call qualty, and so s consdered a second generaton moble phone system. Ths has also meant that data communcaton were bult nto the system usng the 3rd Generaton Partnershp Project [18]. The ubquty of the GSM standard has been advantageous to both consumers and also to network operators, who can choose equpment from any of the many vendors mplementng GSM. GSM also poneered a low-cost alternatve to voce calls, the Short message servce (SMS), whch s now supported on other moble standards as well. Newer versons of the standard were backward-compatble wth the orgnal GSM phones. For example, Release '97 of the standard added packet data capabltes, by means of General Packet Rado Servce (GPRS). Release '99 ntroduced hgher speed data transmsson usng Enhanced Data Rates for GSM Evoluton (EDGE). GSM s a cellular network, whch means that moble phones connect to t by searchng for cells n the mmedate vcnty. GSM networks operate n four dfferent frequency ranges. Most GSM networks operate n the 900 MHz or 1800 MHz bands. Some countres n the Amercas (ncludng Canada and the Unted States) use the 850 MHz and 1900 MHz bands because the 900 and 1800 MHz frequency bands were already allocated. The rarer 400 and 450 MHz frequency bands are assgned n some countres, notably Scandnava, where these frequences were prevously used for frst-generaton systems

151 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION In the 900 MHz band the uplnk frequency band s MHz, and the downlnk frequency band s MHz. Ths 25 MHz bandwdth s subdvded nto 124 carrer frequency channels, each spaced 200 khz apart. Tme dvson multplexng s used to allow eght full-rate or sxteen half-rate speech channels per rado frequency channel. There are eght rado tmeslots (gvng eght burst perods) grouped nto what s called a TDMA frame. Half rate channels use alternate frames n the same tmeslot. The channel data rate s kbt/s, and the frame duraton s ms. The transmsson power n the handset s lmted to a maxmum of 2 watts n GSM850/900 and 1 watt n GSM1800/1900. GSM has used a varety of voce codecs to squeeze 3.1 khz audo nto 5.6 to 13 kbt/s. Orgnally, two codecs, named after the types of data channel they were allocated, were used, called Half Rate (5.6 kbt/s) and Full Rate (13 kbt/s). These used a system based upon lnear predctve codng. In addton to beng effcent wth btrates, these codecs also made t easer to dentfy more mportant parts of the audo, allowng the ar nterface layer to prortze and better protect these parts of the sgnal. GSM was further enhanced n 1997 wth the Enhanced Full Rate (EFR) codec, a 12.2 kbt/s codec that uses a full rate channel. Fnally, wth the development of UMTS, EFR was re-factored nto a varable rate codec called AMR-Narrowband, whch s hgh qualty and robust aganst nterference when used on full rate channels, and less robust but stll relatvely hgh qualty when used n good rado condtons on half-rate channels. There are four dfferent cell szes n a GSM network macro, mcro, pco and umbrella cells. The coverage area of each cell vares accordng to the mplementaton envronment. Macro cells can be regarded as cells where the base staton antenna s nstalled on a mast or a buldng above average roof top level. Mcro cells are cells whose antenna heght s under average roof top level; they are typcally used n urban areas. Pcocells are small cells whose coverage dameter s a few dozen meters; they are manly used ndoors. Umbrella cells are used to cover shadowed regons of smaller cells and fll n gaps n coverage between those cells. Cell horzontal radus vares dependng on antenna heght, antenna gan and propagaton condtons from a couple of hundred meters to several tens of klometers. The longest dstance the GSM specfcaton supports n practcal use s 35 klometres. There are also several mplementatons of the concept of an extended cell, where the cell radus could be double or even more, dependng on the antenna system, the type of terran and the tmng advance. Indoor coverage s also supported by GSM and may be acheved by usng an ndoor pcocell base staton, or an ndoor repeater wth dstrbuted ndoor antennas fed through power spltters, to delver the rado sgnals from an antenna outdoors to the separate ndoor dstrbuted antenna system. These are typcally deployed when a lot of call

152 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION capacty s needed ndoors, for example n shoppng centers or arports. However, ths s not a prerequste, snce ndoor coverage s also provded by n-buldng penetraton of the rado sgnals from nearby cells. The modulaton used n GSM s Gaussan mnmum-shft keyng (GMSK), a knd of contnuous-phase frequency shft keyng. In GMSK, the sgnal to be modulated onto the carrer s frst smoothed wth a Gaussan low-pass flter pror to beng fed to a frequency modulator, whch greatly reduces the nterference to neghbour channels (adjacent channel nterference) The GPRS protocol General Packet Rado Servce (GPRS) s a Moble Data Servce avalable to users of GSM and IS-136 moble phones. It provdes data rates from 56 up to 114 Kbps. GPRS data transfer s typcally charged per klobyte of transferred data, whle data communcaton va tradtonal crcut swtchng s blled per mnute of connecton tme, ndependent of whether the user has actually transferred data or has been n an dle state. GPRS can be used for servces such as Wreless Applcaton Protocol (WAP) access, SMS, MMS and for Internet communcaton servces such as emal and World Wde Web access. Such a technology provdes moderate speed data transfer, by usng unused Tme dvson multple access (TDMA) channels n, for example, the GSM system. Orgnally there was some thought to extend GPRS to cover other standards, but nstead those networks are beng converted to use the GSM standard, so that GSM s the only knd of network where GPRS s n use. GPRS s ntegrated nto GSM Release 97 and newer releases. It was orgnally standardzed by European Telecommuncatons Standards Insttute (ETSI), but now by the 3rd Generaton Partnershp Project. The multple access methods used n GSM wth GPRS are based on frequency dvson duplex and TDMA. Durng a sesson, a user s assgned to one par of up-lnk and downlnk frequency channels. Ths s combned wth tme doman statstcal multplexng,.e. packet mode communcaton, whch makes t possble for several users to share the same frequency channel. The packets have constant length, correspondng to a GSM tme slot. The down-lnk uses frst-come frst-served packet schedulng. GPRS orgnally supported (n theory) Internet Protocol (IP) and Pont-to-Pont Protocol (PPP). GPRS s new technology whch speed s a drect functon of the number of TDMA tme slots assgned, whch s the lesser of what the partcular cell supports and the maxmum capablty of the moble devce expressed as a GPRS Multslot Class. Transfer speed depends also on the channel encodng used. The least robust, but fastest, codng scheme (CS-4) s avalable near a base transcever staton (BTS), whle the most robust

153 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION codng scheme (CS-1) s used when the moble staton s further away from a BTS. Usng the CS-4 t s possble to acheve a user speed of 20.0 kbt/s per tme slot. However, usng ths scheme the cell coverage s 25% of normal. CS-1 can acheve a user speed of only 8.0 kbt/s per tme slot, but has 98% of normal coverage. Newer network equpment can adapt the transfer speed automatcally dependng on the moble locaton. GPRS s packet based. When TCP/IP s used, each phone can have one or more IP addresses allocated. GPRS wll store and forward the IP packets to the phone durng cell handover (when you move from one cell to another). A rado nose nduced pause can be nterpreted by TCP as packet loss, and cause a temporary throttlng n transmsson speed

154 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5.6. Metrologcal characterzaton of the measurement system - The Supplement 1 to the G.U.M. The Supplement to the Gude to the expresson of Uncertanty n Measurement (GUM) s concerned wth the propagaton of probablty dstrbutons through a mathematcal model of measurement as a bass for the evaluaton of uncertanty of measurement, and ts mplementaton by a Monte Carlo method. The treatment apples to a model havng any number of nput quanttes, and a sngle output quantty. The descrbed Monte Carlo method s a practcal alternatve to the GUM uncertanty framework. It has partcular value when: ) lnearzaton of the model provdes an nadequate representaton of the measurement process, or ) the probablty densty functon (PDF) for the output quantty departs apprecably from a Gaussan dstrbuton or a scaled and shfted t-dstrbuton, e.g. due to marked asymmetry. In the former case the estmate of the output quantty and the assocated standard uncertanty provded by the GUM uncertanty framework mght be unrelable. In case ), unrealstc coverage ntervals (a generalzaton of expanded uncertanty n the GUM uncertanty framework) mght be the outcome. The GUM [3] provdes a framework for assessng uncertanty, whch s based on the use of the law of propagaton of uncertanty and the characterzaton of the output quantty by a Gaussan dstrbuton or a scaled and shfted t-dstrbuton. Wthn that framework, the law of propagaton of uncertanty provdes a means for propagatng uncertantes through the model. Specfcally, t evaluates the standard uncertanty assocated wth an estmate of the output quantty, gven a) best estmates of the nput quanttes, b) the standard uncertantes assocated wth these estmates, and, where approprate, c) degrees of freedom assocated wth these standard uncertantes, as well as d) any nonzero covarance assocated wth pars of these estmates. Also wthn the framework, the PDF taken to characterze the output quantty s used to provde a coverage nterval, for a stpulated coverage probablty, for that quantty. The best estmates, standard uncertantes, covarances and degrees of freedom summarze the nformaton avalable concernng the nput quanttes. Wth the approach consdered n the Supplement 1 [4], the avalable nformaton s encoded n terms of PDFs for the nput quanttes. The GUM uncertanty framework does not explctly refer to the assgnment of PDFs to the nput quanttes. However a Type A standard uncertanty s

155 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION obtaned from a probablty densty functon derved from an observed frequency dstrbuton, whle a Type B standard uncertanty s obtaned from an assumed probablty densty functon based on the degree of belef that an event wll occur. Both approaches employ recognzed nterpretatons of probablty. The approach operates wth the PDFs of the nput quanttes n order to determne the PDF for the output quantty. Whereas there are some lmtatons to the GUM uncertanty framework, the propagaton of dstrbutons wll always provde a PDF for the output quantty that s consstent wth the PDFs for the nput quanttes. Ths PDF for the output quantty descrbes the knowledge of that quantty, based on the knowledge of the nput quanttes, as descrbed by the PDFs assgned to them. Once the PDF for the output quantty s avalable, that quantty can be summarzed by a) ts expectaton, taken as an estmate of the quantty, and b) ts standard devaton, taken as the standard uncertanty assocated wth the estmate. Further, the PDF can be used to obtan c) a coverage nterval, correspondng to a stpulated coverage probablty, for the output quantty. The PDF for a quantty expresses the state of knowledge about the quantty,.e. t quantfes the degree of belef about the values that can be assgned to the quantty based on the avalable nformaton. The nformaton usually conssts of raw statstcal data, results of measurement or other relevant scentfc statements. The propagaton of dstrbutons has wder applcaton than the GUM uncertanty framework. It works wth rcher nformaton than that conveyed by best estmates and the assocated standard uncertantes (and degrees of freedom and covarances when approprate). Gven the model relatng the nput quanttes and the output quantty and the PDFs characterzng the nput quanttes, there s a unque PDF for the output quantty. Generally, the latter PDF cannot be determned analytcally. Therefore, the objectve of the approach descrbed n [4] s to determne (a), (b), and (c) above to a prescrbed numercal tolerance. The unqueness of the PDF for the output quantty depends on the model defnng the output quantty unquely n terms of the nput quanttes. In [4] are not consdered models that do not defne the output quantty unquely (for example, nvolvng the soluton of a quadratc equaton, wthout specfyng whch root s to be taken) Man stages of uncertanty evaluaton The man stages of uncertanty evaluaton consttute formulaton, propagaton, and summarzng: a) Formulaton: 1) defne the output quantty, the quantty ntended to be measured (the measurand);

156 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 2) determne the nput quanttes upon whch the output quantty depends; 3) develop a model relatng the output quantty to these nput quanttes; 4) on the bass of avalable knowledge assgn PDFs Gaussan (normal), rectangular (unform), etc. to the nput quanttes. Assgn nstead a jont PDF to those nput quanttes that are not ndependent; b) Propagaton: propagate the PDFs for the nput quanttes through the model to obtan the PDF for the output quantty; c) Summarzng: use the PDF for the output quantty to obtan 1) the expectaton of that quantty, taken as an estmate of the quantty, 2) the standard devaton of that quantty, taken as the standard uncertanty assocated wth the estmate [3], 3) a coverage nterval contanng the output quantty wth a specfed probablty (the coverage probablty) Propagaton of dstrbutons When Y s the result of an ndrect measurement,.e. t s obtaned as a functon of N drect measurements X 1, X 2,, X N, a formal defnton for the PDF for Y s: η) =... g X ( ξ) δ( η f ( ξ)) dξ N... ξ1 gy ( d (5.1) where δ( ) denotes the Drac delta functon. The multple ntegral (5.1) cannot generally be evaluated analytcally. A numercal ntegraton rule can be appled to provde an approxmaton to g Y (η), but ths s not an effcent approach. In [4] a generally effcent approach for determnng (a numercal approxmaton to) the dstrbuton functon G η Y ( η) = gy ( z) dz for Y s consdered. It s based on applyng a Monte Carlo method (MCM) as an mplementaton of the propagaton of dstrbutons. An estmate of Y s the expectaton E(Y). The standard uncertanty assocated wth ths estmate s gven by the standard devaton of Y, the postve square root of the varance V(Y) of Y. A coverage nterval for Y can be determned from G Y (η). Let denote any numercal value between zero and (1 p), where p s the requred coverage probablty. The endponts of a 100p% coverage nterval for Y are 1 ( α G ) and 1 ( p + α).e. the α- and (p+α)- quantles of G Y (η). The choce α=(1 p)/2 gves the coverage nterval defned by the (1 p)/2- and (1+p)/2- quantles, provdng a probablstcally symmetrc 100p % coverage nterval. A numercal value of α dfferent from (1 p)/2 may be more approprate f the PDF s asymmetrc. The shortest 100p% coverage nterval can be used n ths case. It has the property that, for a Y G Y

157 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION unmodal (snglepeaked) PDF, t contans the mode, the most probable value of Y. It s gven by the numercal value of α such that G ( p + α) 1 1 Y G Y ( α) s a mnmum. The probablstcally symmetrc 100p% coverage nterval and the shortest 100p% coverage nterval are dentcal for a symmetrc PDF, such as the Gaussan and scaled and shfted t-dstrbuton used wthn the GUM uncertanty framework [3]. Therefore, n comparng the GUM uncertanty framework wth other approaches, ether of these ntervals can be used. The propagaton of dstrbutons can be mplemented n several ways: a) analytcal methods, provdng a mathematcal representaton of the PDF for Y ; b) uncertanty propagaton based on replacng the model by a frst-order Taylor seres approxmaton the law of propagaton of uncertanty; c) as b), except that contrbutons derved from hgher-order terms n the Taylor seres approxmaton are ncluded; d) numercal methods that mplement the propagaton of dstrbutons, specfcally usng MCM. Analytcal methods are deal n that they do not ntroduce any approxmaton. They are applcable n smple cases only, however. MCM as consdered here s regarded as a means for provdng a numercal representaton of the dstrbuton for the output quantty, rather than a smulaton method per se. In the context of the propagaton stage of uncertanty evaluaton, the problem to be solved s determnstc, there beng no random physcal process to be smulated. Approaches to uncertanty evaluaton other than the GUM uncertanty framework are permtted by the GUM. The approach advocated n [4], based on the propagaton of dstrbutons, s general. For lnear or lnearzed models and nput quanttes for whch the PDFs are Gaussan, the approach yelds results consstent wth the GUM uncertanty framework. However, n cases where the condtons for the GUM uncertanty framework to be appled do not hold, the approach of ths Supplement can generally be expected to lead to a vald uncertanty statement. The condtons necessary for the GUM uncertanty framework to gve vald results hold, then that approach can be used. If there are ndcatons that the GUM uncertanty framework s lkely to be nvald, then another approach should be employed. A thrd stuaton can arse n whch t s dffcult to assess whether or not the GUM uncertanty framework wll be vald. In all three cases, MCM provdes a practcal (alternatve) method. In the frst case, MCM may sometmes be easer to apply due to dffcultes n calculatng senstvty coeffcents, for example. In the second, MCM can generally be expected to gve vald results, snce t does not make approxmatng assumptons. In the thrd, MCM

158 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION can be appled ether to determne the results drectly or to assess the qualty of those provded by the GUM uncertanty framework. Fgure Illustraton of the propagaton of dstrbutons for N=3 ndependent nput quanttes. The propagaton of the PDFs g X (ξ ), =1,, N, for the nput quanttes X through the model to provde the PDF g Y (η) for the output quantty Y s llustrated n Fgure 5-10 for N = 3 ndependent X. The g X (ξ ), = 1, 2, 3, are Gaussan, trangular, and Gaussan, respectvely. g Y (η) s ndcated as beng asymmetrc, as generally arses for non-lnear models or asymmetrc g X (ξ ) Monte Carlo approach to the propagaton MCM provdes a general approach to obtan an approxmate numercal representaton G of the dstrbuton functon G Y (η) for Y. The heart of the approach s repeated samplng of the PDFs for the X and the evaluaton of the model n each case. Snce G Y (η) encodes all the nformaton known about Y, any property of Y such as expectaton, varance and coverage ntervals can be approxmated usng G. The qualty of these calculated results mproves as the number of tmes the PDFs are sampled ncreases. Expectatons and varances (and hgher moments) can be determned drectly from the set of model values obtaned. The determnaton of coverage ntervals requres these model values to be ordered. If y r, r = 1,, M, represent M model values sampled ndependently from a probablty dstrbuton for Y, then the expectaton E(Y) and varance V(Y) can be approxmated usng the y r. In general, the moments of Y (ncludng E(Y) and V(Y)) are approxmated by those of the sampled model values. Let M y0 denote the number of y r that are no greater than y 0, any prescrbed number. The probablty P r (Y<y 0 ) s approxmated by M y0 /M. In ths way, the y r provde a step functon (hstogram-lke) approxmaton to the dstrbuton functon G Y (η). Each y r s obtaned by samplng at random from each of the PDFs for the X and evaluatng the model at the sampled values so obtaned. G, the prmary output from MCM, consttutes the y r arranged n monotoncally ncreasng order

159 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION MCM as an mplementaton of the propagaton of dstrbutons s shown dagrammatcally n Fgure 5-11 for M provded n advance. MCM can be stated as a step-by-step procedure: a) select the number M of Monte Carlo trals to be made; b) generate M vectors, by samplng from the assgned PDFs, as realzatons of the set of N nput quanttes X ; c) for each such vector, form the correspondng model value of Y, yeldng M model values; d) sort these M model values nto non-decreasng order, usng the sorted model values to provde G; e) use G to form an estmate y of Y and the standard uncertanty u(y) assocated wth y; f) use G to form an approprate coverage nterval for Y, for a stpulated coverage probablty p. Mathematcally, the average of the M model values s a realzaton of a random varable wth expectaton E(Y) and varance V(Y)/M. Thus, the closeness of agreement between ths average and E(Y) can be expected to be proportonal to M 1/2. Step e) can equally be carred out by usng the M model values of Y unsorted. It s necessary to sort these model values f the coverage nterval n step f) s requred. The effectveness of MCM to determne y, u(y) and a coverage nterval for Y depends on the use of an adequately large value of M (step a). The propagaton of dstrbutons mplemented usng MCM can valdly be appled, and the requred summary nformaton subsequently determned, n terms of the nformaton provded n [4], under the followng condtons: f s contnuous wth respect to the elements X of X n the neghbourhood of the best estmates x of the X ; no condton on the dervatves of f s requred; the dstrbuton functon for Y s contnuous and strctly ncreasng. These two condtons are necessary to ensure that the nverse of the dstrbuton functon s unque and hence coverage ntervals can be determned. Only the frst condton s needed f a coverage nterval s not requred. the PDF for Y s contnuous over the nterval for whch ths PDF s strctly postve, unmodal (sngle-peaked), strctly ncreasng (or zero) to the left of the mode and strctly decreasng (or zero) to the rght of the mode. Such condton s necessary to ensure that the shortest coverage nterval correspondng to a stpulated coverage probablty s unque. The mode may occur at an endpont of the nterval over whch ths PDF s strctly postve, n whch case one of the two above condtons s vacuous

160 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION E(Y) and V(Y) exst; ths condton s needed for convergence of MCM as the number M of trals ncreases. a suffcently large value of M s used. Fgure The propagaton and summarzng stages of uncertanty evaluaton usng MCM to mplement the propagaton of dstrbutons. When usng the prncple of maxmum entropy, ntroduced by Jaynes [5], a unque PDF s selected among all possble PDFs havng specfed propertes, e.g. specfed central moments of dfferent orders or specfed ntervals for whch the PDF s non-zero. Ths method s partcularly useful for assgnng PDFs to quanttes for whch a seres of ndcatons s not avalable or to quanttes that have not explctly been measured at all. In applyng the prncple of maxmum entropy, to obtan a PDF g X (ξ) that adequately characterzes ncomplete knowledge about a quantty X accordng to the nformaton

161 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION avalable, the functonal S[ g] = g ξ) ln g ( ξ dξ (,.e. the nformaton entropy, s X X ) maxmzed under constrants gven by the nformaton. If the only avalable nformaton regardng a quantty X s a lower lmt a and an upper lmt b wth a<b, then, accordng to the prncple of maxmum entropy, a rectangular dstrbuton R(a, b) over the nterval [a, b] would be assgned to X. the PDF for X s then: g g X X 1 ( ξ) = a ξ b b a (5.2) ( ξ) = 0 otherwse X expectaton and varance are respectvely: a + b E( X ) = (5.3) 2 2 ( b a) V ( X ) = (5.4) 12 Table 5-3 reports the crtera suggested by the Gude to assgn a certan PDF to each nput quantty X to be consdered for usng the Monte Carlo approach to estmate the PDF of Y and hence the combned standard uncertanty on t

162 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Table 5-3. PDF assgned on the bass of avalable nformaton

163 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5.7. Evaluaton of the combned uncertanty on the poston of the transent source To complete an exhaustve analyss of the proposed method, the measurement system needs to be characterzed n terms of uncertanty on the poston of the fault. Hence, each block of a voltage channel of a remote unt (slave) s characterzed to evaluate all the contrbutons to the uncertanty on the captured tme stamp. In fact, the uncertanty on the poston of the fault depends on the uncertanty affectng the tme stamps captured by the slaves and then processed by the master unt. The propagaton of the uncertanty source effects through the fault locaton algorthm has been analyzed. The combned standard uncertanty u(t j ) on the tme stamp t j s due to the random effects of the uncertanty sources located n all the devces of the measurement hardware, whereas the bases are assumed to be known; hence, compensated. Expermental tests for the metrologcal characterzatons of the VVT, the EDB and the GPS staton have been carred out: the effects of the uncertanty sources located n the measurement chan have been obtaned and expressed as PDFs; then, a numerc technque,.e. the Monte Carlo Smulaton (MCS) procedure, has been run to evaluate u(t j ). In ths connecton, two assumptons were taken: ) the statstcal mean value of the random varable s the bas affectng t j tself; ) all the tme stamps that can be captured by any channel of any slave have smlar frequency dstrbuton. The MCS procedure s appled to phase-to-ground voltage sgnals prevously obtaned by the EMTP-RV smulatons descrbed n Secton 4.2; the voltage montored n each node of the consdered network n dfferent fault cases s used as nput. The accuracy performance of the fault locaton method has been tested n realstc workng condtons Metrologcal Characterzaton of the voltage transducer The voltage transducer chosen for the characterzaton s a capactve dvder used for AC hgh-voltages by the man Italan electrcal utlty. Its nomnal specfcatons, n ol and nto 1-MΩ load, are: maxmum pulse voltage 300 kv; voltage dvson rato 3800:1; bandwdth 30Hz 4MHz; droop rate 0.02%/μs; usable rse-tme 100 ns. The transducer transfer functon has been determned to buld up the equvalent model of the condtonng block used n the MCS procedure. In the experments, both magntude- and phase- frequency response have been obtaned by processng the data smultaneously acqured at the nput and output of the VVT, as shown n Fgure The calbrator Wavetek Datron 4800A has been used to feed the VVT. It can provde up to 1000 V for frequences up to 33 khz, 100 V for frequences up to 100 khz and 10 V up to 1 MHz. Its nomnal accuracy specfcatons are: 50 ppm for AC voltage; zero-to-full-range lnearty < 0.1 ppm of full scale. The calbrator has been confgured to generate a snusodal waveform n the range 50 Hz - 1 MHz. In dong ths, the transducer parameters are assumed to depend only on

164 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION frequency and not on prmary voltage. For each frequency, 100 measurements of the RMS of both the nput and output sgnals were taken along wth ther phase shft. IEEE 488 bus DSO DMM n HI DMM out calbrator LO VVT out gnd Fgure Test bank for the VVT characterzaton. Table 5-4. Expected value and standard uncertanty of K and Δφ of the VVT n functon of frequency. freq [Hz] μ(k) E E E E E E E+03 σ(k) E E E-01 μ(δφ) E E E E E E E-02 σ(δφ) E E E E E E E-03 freq [Hz] μ(k) σ(k) μ(δφ) σ(δφ) E E E E E E E E E E E E E E E E E E E E E E E E-02 Ths way, the probablty densty functons of the random varables representng the gan K and the phase error Δφ of the transducer have been determned vs. frequency. For the

165 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION above tests, two hghly-accurate Dgtal Mult Meters (DMMs) HP3458A are used for RMS measurements of the VVT nput and output sgnals. The HP3458 model has been wdely used as reference samplng devce, see e.g. [6, 7]; despte, ts use for computng the sgnals RMS and phase-shft s not possble gven that ts maxmum samplng rate does not ft the frequency range of nterest. Ths DMM supports three technques for RMS measurement, each offerng specfc capabltes: a) synchronously sub-sampled computed RMS technque, b) analog computng RMS converson technque, c) random sampled computed RMS technque. The a) technque features very hgh lnearty and the hghest accuracy; t requres a repettve nput sgnal n the bandwdth 1 Hz 10 MHz. The b) mode s the default one and ensures the fastest measurement speed n the range 10 Hz 2 MHz. Fnally, the c) soluton features hgh lnearty as a) but the lowest accuracy n the bandwdth 20 Hz 10 MHz. Beng the calbrator output repettve and stable, the a) mode seems to be the most sutable for our purpose. The tme delay ntroduced by the VVT vs. frequency s evaluated by acqurng the nput and output sgnals wth a Dgtal Storage Osclloscope (DSO) whose channels feature nomnal smultaneous acquston wth hgh samplng rate. The delay between the nput and output sgnals can be drectly measured by the DSO as tme skew between two nput channels. Ths procedure s automatcally run by controllng the nstruments through a PC va an IEEE 488-based nterface. The measurement results are stored n the PC and then processed to obtan the frequency dstrbutons of the ampltude- and phase- transfer functons. μ(k) frequency 6 Fgure 5-13 Expected value of K of the VVT n functon of frequency (red curve = best ft to expermental results) In Fgure 5-13 and Fgure 5-14 are reported the mean value of the gan K and of the phase error Δφ, respectvely, n functon of frequency. The red curves are obtaned by

166 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION fttng expermental results wth polynomal nterpolatng expressons. In ths way μ(k) and μ(δφ) can be consdered contnuous functons n the frequency doman n the whole range of nterest for the characterzaton. The results on whose bass the curves are obtaned are reported n Table 5-4. μ(δφ) frequency Fgure 5-14 Expected value of Δφ of the VVT n functon of frequency (red curve = best ft to expermental results) Metrologcal characterzaton of the Event Detecton Block IEEE 488 bus DSO u lne u sne + EDB gnd gnd u out u pulse Fgure 5-15 Automatc test bench for the EDB characterzaton The uncertanty contrbuton of the EDB to t j has to be evaluated as the tme delay between the occurrence of the transent and the correspondng change of the logc level of the EDB output. To ths purpose, an automatc test procedure has been developed by

167 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION usng two functon generators and the DSO (see Fgure 5-15). One of the sources s confgured to generate a snusodal waveform u sne at power frequency wth ampltude 10 V. The second functon generator s confgured to gve a 1 V-peak pulse waveform u pulse. The rsng-front tme and the duraton of u pulse are randomly extracted n the ntervals [5 ns 100 ns] and [500 ns 10 us]. The former parameter corresponds to settng the dervatve of the frst surge of the event, whle the latter parameter s equvalent to half perod of the oscllatons n case of oscllatory transent, or to the duraton of the surge n case of mpulsve transent. Ths way, the frequency content of the sgnal s dfferent n each test and ranges through the entre frequency range assocated to EM transent dsturbances affectng power networks. μ(t d) = 9.37 E-7 σ(t d) = 1.2 E-8 Fgure 5-16 Frequency dstrbuton of T d ntroduced by the EDB The sgnal u lne at the nput of the EDB has been obtaned by addng u sne and u pulse. For each value of the rsng tme of u pulse, the DSO acqures u lne and u out to measure the tme skew between the begnnng of the rsng front of the former sgnal and the commutaton from hgh to low level of the latter one. Ten thousand measurements of the tme skew have been carred out to get a probablty dstrbuton featurng a standard devaton that reasonably takes nto account all the combnaton of rsng fronts and duratons that can be found n transents scenaro. The results are reported n Fgure 5-16 by means of a 30 classes hstogram

168 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Metrologcal characterzaton of the GPS staton The GPS staton s the GPS168 PCI by Menberg, whose nomnal accuracy specfcaton are: tme resoluton equal to 100 ns; standard uncertanty on the captured tme lower than 250 ns. In ths case, only a type-b uncertanty evaluaton can be done. Therefore, accordng to [3], a unform probablty densty functon wth zero expected value and 250 ns standard devaton s assumed to represent the GPS contrbuton to u(t j ) Monte Carlo Method to evaluate the combned uncertanty on the fault locaton The results of the metrologcal characterzaton carred out for the blocks of the measurement system are used to evaluate the uncertanty affectng the poston of a fault occurrng wthn the power network montored by the system tself. Accordng to the above descrbed probablstc approach, the procedure takes nto account both bas and random uncertanty contrbutons of the devces, and corrupts the nput voltage of the emulated slave unt by applyng ther effects to each acqured sample. In practce a Matlab scrpt loads as nput the set of voltage waveforms obtaned n correspondence of all the slave statons present n the smulated dstrbuton network when one partcular case of short crcut occurs. The magntude and phase spectrum of each lne-voltage are corrupted by means of the characterstcs of gan K and phase error Δφ, respectvely; every frequency component s dstorted n ampltude, under the effect of both the bas μ(k), (see Fgure 5-13) and a random contrbuton to K assumed to have a Gaussan PDF wth mean value μ(k) and as σ(k) the maxmum value of the expermental standard devaton,.e. the one measured at f = 500 khz (see Table 5-4). The same happens to the phase of the frequency component, affected by both the bas contrbuton μ(δφ) (see Fgure 5-14) and a random varable havng unform PDF, expected value μ(δφ) and ampltude 2 tmes the σ(δφ) measured at f = 500 khz (see Table 5-4). In ths way the worst operatng condtons are assumed for the VVT. The frequency spectrum of the nput voltage fltered by the VVT model s then brought back n the tme doman, so that the transduced sgnal s obtaned. Ths s the waveform that n each slave s gven as nput to the EDB. The Event Detecton Block descrbed n Secton 5.2 has been reproduced n Matlab, where the flters used to obtan u f have been mplemented n the tme doman. The logc sgnal at the output of the EDB functon changes ts level when the dfference between the transduced voltage waveform and ts low frequency content s greater than the EDB threshold. The tmestamp correspondng to the commutaton s then corrupted by addng the delay measured durng the characterzaton of the EDB (see Fgure 5-16), assumed as a random varable T d havng a Gaussan PDF wth μ(t d )=9.37e-7 and σ(t d )=1.2e

169 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Fnally, the effect of the uncertanty of the GPS staton s taken nto account, and to each tmestamp obtaned by the prevous step a random contrbuton s superposed, accordng to what descrbed above Expermental results The MCM descrbed above has been appled to the set of voltage waveforms obtaned at the end of an EMTP smulaton carred out n the dstrbuton network descrbed n Secton 4.2. In partcular, the consdered case of fault was a phase a-to-ground short crcut affectng lne L6 and dstant 2000 m from node S7 (see Fgure 4.8). One thousand MC trals have been run to obtan the set of results used for the evaluaton of the combned uncertanty on the fault locaton. At the end of each tral the nformaton relevant to the lne assumed as faulted, the slave node consdered the nearest one to the fault pont and the estmated dstance of the fault poston from the latter node were saved. In 100% of the trals run the faulted lne and the nearest node to fault have been correctly dentfed as L6 and S7, respectvely. Fgure 5-17 Hstogram of the fault locaton method results n case of lne-to-ground short crcut The frequency dstrbuton of the dstance d computed to locate the short crcut along the lne at the end of each tral s represented n Fgure 5-17 by means of a 30 classes hstogram. The expected value of d corresponds to 1920 m, affected by an error of 80 m respect to the actual fault poston, whereas the standard uncertanty on d s equal to 15.2 m. It should be noted that, accordng to [3], the bas uncertanty contrbuton, once t s

170 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION known, has to be compensated to carry out the characterzaton of the system, hence the fnal uncertanty on the fault locaton can be assocated to the only random contrbuton. A second MCM procedure has been run to evaluate to dsperson of results n the case of phase a to phase c short crcut affectng lne L8, 1 km dstant from slave S9. The results of the 1000 trals are reported n Fgure 5-18 n terms of frequency dstrbuton of the estmated dstance d of the fault pont from node S9. Even n ths case an hstogram featurng 30 classes has been created. The expected value of d corresponds to m, hence t s affected by a bas of 13 m respect to the actual value, and the standard uncertanty s equal to 13.2 m, just a few lower than the standard devaton obtaned n the prevous fault case. The dfferent accuracy affectng the fault locaton n the two analyzed cases turns nto a dfferent polarzaton of the frequency dstrbutons; the expected value of the dstance d s affected by a greater bas n the case of grounded short crcut respect to the ungrounded one maybe because the propagaton speed assumed by the master for the travellng waves s more correct n the latter fault condton than n the former one. Moreover, a grounded fault s usually less heavy for the power system than a fault between phase conductors, and thus the surge rsng at the fault pont and propagatng through the lnes s more dumped. By consderng the applcaton of the fault locaton procedure, as already explaned n prevous sectons, ths fact leads to a more dffcult and consequently less accurate locaton of the fault. Fgure 5-18 Hstogram of the fault locaton method results n case of lne-to-lne short crcut

171 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5A. Appendx: Uncertanty Contrbuton of the Analog Condtonng Block n DSP-based Instruments The type-b evaluaton of the uncertanty affectng the measurements provded by any nstrument could be done by properly processng the nomnal accuracy specfcatons of each functonal block n the nstrument. In DSP-based measurements, the mplementaton of such a procedure faces the dffculty arsng from the fact that the usual accuracy specfcatons of the nput analog condtonng block cannot be used to determne the uncertanty affectng each sample of the block output sgnal. A calbraton procedure has been developed to try to tackle ths ssue. It s based on the use of hghaccuracy data acquston boards and the smultaneous acquston of the sgnals at the nput and output of the analog block. Bascally, any measurng nstrument based on Dgtal Sgnal Processng (DSP) technques conssts of three blocks: an nput block, whch contans transducers and analog condtonng crcuts; a data acquston and A/D converson (DAQ) block; a control and data processng block. The uncertanty estmaton n DSP-based measurements requres that the uncertanty affectng each acqured data s known. A problem mmedately arses n ths estmaton, because of the dfferent characterzaton procedures of the analog condtonng blocks and DAQ boards. As a matter of fact, the accuracy specfcatons of a DAQ board are defned wth reference to a generc sample, whereas the parameters gven to specfy the accuracy of the nput analog block devces do not allow the estmate of the uncertanty affectng the samples of the block output. For nstance, the tradtonal nstrument transformers are characterzed by means of the rato and phase angle errors (see, e.g. the Internatonal Standards [8, 9]). Even n the case of electronc voltage transformers whch feature a dgtal output secton, no parameter characterzng the accuracy of the samples s defned by IEC Std [10]. The method for the characterzaton of the analog nput blocks, whch allows determnng the uncertanty affectng each sample of the block output, s based on the comparson between samples smultaneously acqured at the block nput and output by means of hghly-accurate acquston and A/D converson systems. Such a procedure provdes a calbraton curve whch takes nto account all the effects of the uncertanty sources located n the block. The nformaton derved from the curve can then be used, along wth the accuracy specfcatons of the DAQ block, to evaluate the uncertanty affectng the estmate of the measurand. The method could be used by the manufacturer to test a sgnfcant statstcal sample of a gven block model, n order to provde accuracy specfcatons characterzng each output sample of any devce of that type. In the case of voltage transducers, the method can be mplemented by means of the measurement setup descrbed n the followng

172 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5A.1. The calbraton method The method can be mplemented under both AC and DC condtons, and allows the estmate of the uncertanty affectng each output sample of the nstrument analog condtonng block. (a) u n [V] samples (b) u out [V] samples Fgure 5-19 Some of the correspondng samples of the nput and output waveforms used to buld up the calbraton curve Let us assume, only by way of example, that the AC sgnal s snusodal and the nput block s a voltage-to-voltage transducer. Moreover, let us refer to Fgure 5-19, where both the nput and output voltages denoted by u n (t) and u out (t), respectvely, are smultaneously acqured over N p perods, thus obtanng two sequences of length N s. Let us denote ther generc elements by u n [n] and u out [n], respectvely. As for the example n the above fgure, t s N p = 5, N s = 150. In both Fgure 5-19 and what follows, the output sgnal s multpled by the rato R U between the transducer nomnal nput and output. Hence, the waveforms only dffer because of the transducer uncertanty. The fgure qualtatvely shows the process of buldng up the two above sequences. Of course, due to the plot scale, Fgure 5-19 cannot show the dfferences between the nput and output sgnals; therefore, the two graphs seem dentcal. Two perodc sequences of length N s /N p are then bult up. Ther generc element s denoted by u [ ] and u [ ], respectvely. They are defned as follows: n n out n

173 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION N 1 1 p N s n [ n] = un[ n + k ] N p k= 0 N p u (5A.1) N 1 1 p N s out [ n] = RU uout[ n + k ] N p k = 0 N p u (5A.2).e., they are the mean value of the samples havng the same poston n each perod of the observaton nterval. Ths averagng procedure reduces the random effects that could affect the data acqured. As for the tests under DC condtons, a sequence of DC sgnals s appled and ther ampltude s ncreased up to the transducer under test (TUT) full scale; the TUT nput and output are acqured and averaged. Both n AC and DC condtons, the averaged samples of the nput and output sgnals are the x-doman and y-doman, respectvely, of the calbraton curve. The curve takes nto account all the effects of the uncertanty sources located n the TUT; then a lnear fttng process s used to obtan a sutable best ft straght lne (BSL). Three well-known lnearzaton crtera are exemplfed n Fgure 5-20: (a) the BSL s obtaned by means of a lnear regresson technque, whch determnes slope and offset of the lne tself; (b) the lnear regresson technque s appled to get a BSL through zero (.e. wthout offset); (c) the BSL through the end ponts of the calbraton curve s consdered. After that, the devce can be characterzed by means of ndexes derved from: the comparson between the slopes of the BSL and the deal characterstc of the devce tself; the worst devaton of the calbraton curve from the BSL; the BSL offset (f any). 5A.2. Characterzaton ndces Two ndces are used to characterze the devaton of the actual behavour of the TUT from the deal one. α and β are defned as follows: α = Δ β= Δ g g g max n n { u out[ n] g u n[ n] } max{ u out[ n] } (5A.3) (5A.4) In (5A.3) and (5A.4) denotes the absolute value of. In (5A.3), g s the angular coeffcent of the BSL and s computed wth reference to Fgure 5-20 b by applyng the least square method, whereas g n (whch s unty n sutable coordnates) s the slope of the deal characterstc. Hence, α s a gan error. As for β, t s smlar to the ntegral nonlnearty of a DAC, as endorsed by [11]

174 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Fgure 5-20 Lnearzaton crtera If the devce can be consdered lnear, the two ndexes can be used also n nonsnusodal condtons. The procedure proposed could be dffcult to mplement n common laboratores (where hgh-accuracy samplng devces may not be avalable), but can easly be mplemented n a manufacture laboratory. Moreover, f a sgnfcant statstcal sample of a gven transducer model s characterzed accordng to ths procedure, the maxmum values A U-U and B U-U found for α and β, could then be gven n the data sheet to qualfy that transducer model

175 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION Trgger Sgnal V DAQ n u n U-U u out DAQ out IEEE 488 IEEE 488 Fgure 5-21 Block dagram of the calbraton setup Fgure 5-21 s the schematc block dagram of the setup used to characterze the TUT by means of the proposed procedure. U-U denotes the voltage-to-voltage transducer under test, V refers to a sutable reference voltage source (whch must feature hgh short-term stablty) and DAQ refers to a hgh-accuracy data acquston devce. Two HP 3458A have been used as hgh-accuracy samplng devces. HP 3458A features 0.02% of readng, plus an offset voltage varyng from 8 mv at 10 V to 0.8 V at 1 kv. Table 5-5 Values of A U-U and B U-U, along wth ther relevant standard uncertantes u(a U-U ) and u(b U-U ), respectvely, vs. frequency (nput voltage RMS value: 250 V) Frequency [Hz] DC A U-U u(a U-U ) B U-U u(b U-U ) The method requres the smultaneous samplng of u n and u out, otherwse the contrbuton of the transducer delay to α and β gets lost. To ths purpose, the A/D converson start of DAQ n has been used to trgger the operaton of DAQ out. Actve transducers LEM LV-25P have been characterzed by means of the above setup. They bascally are current transducers, but behave as voltage transducers by addng sutable resstors n seres wth ther nput and output. The tested transducers feature the followng nomnal specfcatons: maxmum nput voltage: 400 V; converson voltage rato R U = 40.7 : 1; overall accuracy: 0.8% at full scale; offset voltage: 30 mv. In the DC tests, nput sgnals havng ampltude varyng n steps of 50 V from -250 V to 250 V have been appled. As for the AC tests, the RMS value of the snusodal nput has been kept constant at 250 V, whereas the sgnal frequency has been vared n the nterval [50 Hz, 3 khz]. The maxmum frequency has been chosen accordng to the requrements

176 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION of the IEEE Standard [12] for harmoncs evaluaton. The above frequency range also contans the one endorsed by the European Standard EN [13]. Table 5-5 refers to the characterzaton of three voltage transducers LEM LV 25-P and reports, vs. some test frequences, the maxmum absolute values A U-U and B U-U characterzng the consdered set, along wth the relevant standard uncertanty. 5A.3. Applcaton example The purpose of ths applcaton example s verfyng whether the above ndexes A U-U and B U-U, taken to characterze the consdered transducer set, can be used to evaluate the uncertanty affectng a parameter measured by a DSP-based measurement chan contanng any transducer belongng to the set. In affrmatve case, A U-U and B U-U could be determned, and then reported n the transducer data sheet, by testng a sgnfcant statstcal sample of them n the manufacture laboratory. The problem has been seen by the pont of vew of a transducer manufacturer on the one hand and of a user on the other hand, under the assumpton that they bult up two smlar measurement chans based on a voltage transducer and a DAQ board for the measurement of a RMS voltage. The manufacturer s assumed to have already characterzed a sgnfcant statstcal sample of transducers of the same type used n the chan, by means of the descrbed procedure. He would lke to verfy the usefulness of the ndexes obtaned. As for the user, he wants to determne the uncertanty affectng the RMS measurement. As explaned n chapter 5, the uncertanty can be estmated by applyng the GUM [3] or other approaches that can be found n the lterature, see e.g. [14-16]. When the measurand s not drectly measured, the GUM endorses the Law of Propagaton of Uncertanty whch reles on the lnearzaton of the functonal relatonshp between the measurand and the nput quanttes. Recently, the ISO document [4] has endorsed the use of a MCM to evaluate the uncertanty by means of numercal methods for the propagaton of probablty dstrbutons through the measurement process. HA DAQ U CAL CAL U-U DAQ PC Measurement chan Fgure 5-22 Block dagram of the setup used for the applcaton example Fgure 5-22 shows the block dagram of the setup mplemented to run the applcaton example. CAL refers to the reference voltage source Wavetek 4800, whose nomnal

177 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION accuracy, n terms of expanded uncertanty wth 95%-confdence level, s 40 ppm. The measurement chan conssts of a transducer LEM LV 25-P and a DAQ board NI PCI 6070 havng the followng nomnal specfcatons [17]: offset O DAQ = 6.38 mv; percentage of readng α DAQ = %; nose + quantzaton Q DAQ = 6.10 mv. A Personal Computer (PC) controls both the setup operaton and provdes estmates of the measurand. It also performs offlne estmates of the relevant uncertantes by processng the data acqured by the DAQ board and the hgh-accuracy samplng devce HA DAQ, along wth the parameters characterzng the accuracy of the data at the output of each devce of the measurement chan. The tests have been carred out n snusodal condtons by settng the calbrator output u CAL (t) at the RMS voltage U CAL = 230 V, wth frequency f = 50 Hz. Two MCS procedures have therefore been mplemented, accordng to the ISO document [4]. In both cases the acqured data have been corrupted by the effects, treated as random varables, of the uncertanty sources located n U-U and the DAQ board. Then, the data have been processed to obtan an estmate of the measurand and of ts type-b uncertanty, whch, n accordance wth [4], has been expressed as the shortest coverage nterval assocated to the confdence level of 95%. In any case, let us denote by α U and β U, respectvely, the random varables characterzng the effects of the uncertanty sources located n the voltage transducer. They have unform dstrbuton wthn the ntervals [ A U, + A U ] and [ B U,+ B U ]. Moreover, as for the DAQ board, let us denote by o DAQ, α DAQ, q DAQ the random varables relevant to the uncertanty sources located n the board tself. They have unform dstrbuton wthn the ntervals [ O DAQ, +O DAQ ], [ A DAQ, +A DAQ ], [ Q DAQ, +Q DAQ ], respectvely. In both seres of smulatons, the values reported n Table 5.5 and relevant to 50 Hz,.e. A U = and B U = , have been consdered for the transducers, whereas for the DAQ board the values of O DAQ, A DAQ and Q DAQ found n the data-sheet (reported above) have been used. It s worthwhle hghlghtng that all the above random varables show dfferent statstcal propertes. Among them, q DAQ and β U, are assumed to be totally uncorrelated; the other varables are assumed to be totally correlated. When dfferent measurements (or trals, n the case of smulatons) are performed, the values of the latter random varables are constant n a sngle tral, but randomly vary from one tral to another

178 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 0.03 Number of occurences U DAQ [V] Fgure 5-23 Frequency dstrbuton of the U DAQ values provded by the Monte Carlo smulaton procedure. Now, consder the problem as f t should be tackled by the user, assumed to be provded by the transducer manufacturer of the values of A U and B U. The N-length sequence acqured by the DAQ board presents u[n] as generc element; fnally the MCS procedure s mplemented. The RMS value U DAQm measured n the m-th tral (1 m M) of the MCS has been computed as follows: U DAQ m N 1 = N n= { u[ n] + α u[ n] + β max{ u[ n] } + α u[ n + q + o } R Um U DAQ ] m DAQ DAQm U (5A.5) M = 10 4 trals have been carred out; the dstrbuton n Fgure 5-23 of the obtaned values ~ of U DAQm has been found to be symmetrcal. The statstcal mean value U of the dstrbuton, whch s the estmate of the expected value of the measurand, along wth the estmate of the relevant uncertanty, can therefore be derved at the end of the MCS procedure. The RMS values: U CAL set by the voltage reference source; U DAQ measured by ~ the chan n Fgure 5-22; and U DAQ provded by the MCS procedure have been reported ~ n Fg. 6. Both U DAQ and U DAQ are centred n the nterval representng the uncertanty, ~ and there s compatblty between U DAQ and U CAL (whose 95%-confdence level s plot). DAQ

179 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION U CAL = V V V U DAQ = V V V ~ U = V DAQ ~ U = V CAL V V Fgure 5-24 Measurement results (not n scale). U DAQ : voltage RMS measured by the chan n Fgure 5-22; U CAL : voltage reference source RMS, wth ts 95%-confdence nterval; U ~ and DAQ ~ U : statstcal mean RMS values gauged by applyng the MCS procedures to DAQ- and HA CAL DAQ-output data, respectvely, wth ther 95%-coverage ntervals. If the dgtal sgnal at the DAQ output s seen as the board readng, the evaluaton procedure of the type-b uncertanty affectng U DAQ,.e. processng the data corrupted by the effects of the uncertanty sources located n U-U and DAQ, s qute smlar to the one usually performed to estmate the uncertanty affectng an nstrument readng by applyng the accuracy specfcaton to the readng tself. Another seres of M = 10 4 trals (whch could easly be mplemented n a manufacture laboratory, where hgh accuracy samplng devces are avalable) have been run by processng the data acqured by HA DAQ corrupted by the effects of the uncertanty sources located n U-U and the DAQ board n order to estmate the expected value of ~ ~ U CAL, whch we denote by U CAL, and the relevant uncertanty. U CAL represents another measurand estmate. The effects of the uncertanty sources n HA DAQ have reasonably been neglected. Also ths procedure has provded a symmetrcal dstrbuton of RMS values. The resultng statstcal mean value, along wth the relevant uncertanty, has been reported n Fgure 5-24 too, where t can be seen that there s compatblty between ~ U and the other measurand estmates. The coverage nterval, whch n accordance CAL wth [4] represents the uncertanty, s not centred on U DAQ ; ths s due to the fact that the bas ntroduced by HA DAQ s neglgble f compared to the one caused by the measurement chan. Therefore the ndexes derved by the proposed characterzaton procedure seems to be useful for the contrbuton estmaton of that knd of voltage transducer to the uncertanty n the measurement of U DAQ

180 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 5A.4. An Equpment for Voltage Transducers Calbraton The above method has been mplemented n the case of voltage transducers. The ndexes (5A.3) and (5A.4) are computed for more than one transducer belongng to the same model, obtanng ntervals of values for both of them. If the number of transducers under test corresponds to a consstent statstcal sample, the obtaned calbraton results can be extended to any same-model transducer. Processng Unt Calbraton Results DAQ DAQ Trgger Sgnal generator TUT Fgure 5-25 Block dagram descrbng the proposed method In Fgure 5-25 the calbraton procedure s recalled. A hgh-stablty generator apples the sgnal to the Transducer Under Test (TUT). The method reles on the comparson between sgnals smultaneously acqured at the nput and output of the TUT; hence, an external sgnal s used to trgger the samplng process of two data acquston boards (DAQ). The uncertanty affectng the estmates of α and β, whch are mostly due to the DAQ contrbuton, need to be at least one order of magntude lower than the α and β values. For DC tests, an ncreasng sequence of DC sgnals up to the TUT full scale s appled, and the TUT nput and output are acqured. As for AC condtons, a sutable number N p of perods of both the nput and output quanttes are acqured, then two perodc sequences of length N s, whose generc elements are denoted by x[] and y[] respectvely, are bult up. Fnally, a proper averagng procedure s appled to reduce the random effects on the above elements. Two sequences of length N s /N p are then obtaned, whose generc elements x [] and y [] are defned as follows: N p 1 p k = 0 = Δ 1 N s x [ ] x[ + k ] (5A.6) N N N p 1 p k = 0 p = Δ 1 N s y [ ] y[ + k ] (5A.7) N N p

181 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION The calbraton curve relates the two sequences. A regresson technque s used to ft the curve wth a straght lne forced to cross the axes orgn; fnally the ndexes α and β are computed. When the TUT s a voltage transducer, the block dagram n Fgure 5-25 turns nto the crcut n Fgure U-U denotes the voltage-to-voltage transducer under test, V refers to a sutable voltage source and DAQ n and DAQ out refer to dgtal multmeters (DMM) used as hgh-accuracy data acquston boards. The calbrator Wavetek Datron 4800 has been used as voltage source. It can provde 1000 V at 33 khz and ts nomnal accuracy specfcatons are: 6 ppm for DC voltage; 50 ppm for AC voltage. The zero-to-full-range lnearty s lower than 0.1 ppm of full scale. Two HP3458A have been used as DAQ boards. Trgger Sgnal V DAQ n u n U-U u out DAQ out IEEE 488 IEEE 488 Fgure 5-26 Schematc block dagram of the calbraton setup n case of voltage transducers Each DMM s controlled by a personal computer (PC) va an IEEE 488-based nterface. The dgtal sgnals are stored, n 16-bt format, n the PC n order to be processed. The storng format s compatble wth a lmt of the protocol used. The DMMs can feature two real-tme samplng procedures: drect samplng and DCV dgtzng. Drect-samplng operaton s characterzed by a maxmum samplng rate of 50 ksa/s wth 16-bt resoluton; the nput range can vary from 10 mv to 1000 V. In case of drect samplng the nomnal best accuracy s 0.02% of readng and the offset voltage s 8 mv and 800 mv n the 10Vrange and 1000V-range, respectvely, whch are the ones used n the experments. The nomnal accuracy specfcatons of DCV dgtzng are: 0.005% of readng; offset voltage 500 µv n the 1000-V range and 5µV n the 10V-range. The tme jtter on the samplng perod s lower than 50 ps for both confguratons. DCV dgtzng features lower nose level, hgher resoluton and double maxmum samplng rate (100 ksa/s) respect to drect samplng. Its drawbacks are: greater jtter on the trgger arm and lower nput bandwdth (30 khz vs. 2 MHz for drect-samplng mode). Drect-samplng mode has been preferred

182 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION due to the followng two man reasons: () the better resoluton of DCV s lost by storng the data n 16-bt format; () the bandwdth of 30 khz dd not ft the tests performed. A master-slave confguraton has been adopted to grant a nomnally smultaneous samplng of the nput and output waveforms durng the tests. Fgure 5-26 shows that DAQ n s set as master, whereas DAQ out s set as slave. Thus, the acquston process of DAQ out starts smultaneously wth the A/D converson of DAQ n. To do ths, DAQ n s programmed to trgger DAQ out when the A/D converson starts. Of course, the DAQ out trgger s set as external, whereas the DAQ n one s set as nternal and level (zero crossng). Ths way, the smultaneous acquston s ndependent of the delay ntroduced by the analog condtonng blocks of both nstruments. The nomnal trgger latency,.e. the tme delay between the trgger and the begnnng of the measurement, s specfed to be lower than 125 ns for an external trgger and lower than 700 ns for zero crossng; the nomnal jtter standard devaton s lower than 2 ns. A tme skew between the startng nstant of the acquston at the two sdes of the TUT s expected to occur due to the followng reasons: () dfferent DMMs feature dfferent latences; () there s a tme delay between the begnnng of the A/D converson of the master and the output of the external trgger sgnal; () a tme delay occurs between the trgger-n and the start of the acquston performed by the slave. The tme skew has been measured accordng to the procedure descrbed n what follows. Voltage transducers LEM CV have been calbrated by means of the equpment mplemented. They feature the followng nomnal specfcatons: 1000-V maxmum nput voltage; 500-kHz bandwdth (-1 db); 1000 V : 10 V converson rato; accuracy (referred to as overall accuracy n the data sheets) 0.2% at full scale; 5-mV offset. 5A.5. Uncertanty sources n the equpment V DAQn IEEE 488 DAQout Trgger sgnal Fgure 5-27 Crcut for the measurement of the tme delay T d

183 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION The smultaneous samplng of the TUT nput and output sgnals s the man task n the proposed calbraton procedure, gven that the acquston processes are only nomnally smultaneously performed by the two DAQs. A tme delay T d between ther startng nstants has been evaluated by means of the system n Fgure 5-27 by applyng the same snusodal sgnal to both the samplng devces, whch are connected and controlled as descrbed n the prevous Secton. Dfferent values of the sgnal ampltude and frequency, along wth dfferent nput ranges of the DAQs, have been used n the tests. 30 tests have been performed for each test condton n order to understand whether there s a relatonshp between T d and the above parameters. The value of T d has been determned by measurng the phase shft between the two acqured snusodal waveforms. Table 5-6 reports both ts statstcal mean value µ(t d ) and the relevant samplng standard devaton s(t d ) vs. the test frequences. The values n Table 5-6 have been confrmed by the results of tests performed at dfferent nput ranges and not reported here just for the sake of smplcty. They lead to conclude that T d can be assumed constant and equal to 12 µs, snce n all cases s(t d ) s at least two order of magntude lower than µ(t d ). Hence, µ(t d ) s a bas; t has been corrected by means of an ad-hoc algorthm before evaluatng α and β. Of course, the uncertanty u(t d ) = s(t d ) n the bas estmaton also affects α and β. Table 5-6 µ(t d ) and s(t d ) for some of the dfferent test condtons Frequency [Hz] µ(t d ) [s] s(t d ) [s] Now, let us analyze the effects of the uncertanty sources located n the hardware of the calbraton equpment and how ther effects propagate n the algorthm used to compute α and β. Ths way, the standard uncertantes u(α) and u(β) can be estmated. In addton to u(t d ), the followng effects of the other uncertanty sources have to be analyzed: - the offset of both the DAQs, whose values we denote by O n and O out, respectvely; - the percentage of readng of DAQ n as well as DAQ out, whose values we denote by R n and R out, respectvely; - the tme jtter on the samplng perod. A Monte-Carlo smulaton procedure consstng of 10,000 trals has been performed to evaluate the expanded uncertanty on α and β. Such parameter s expressed as the shortest coverage nterval correspondng to the confdence level of 95% [4]. The measurement algorthm evaluates n each tral: a) the average perods of both nput and output sgnals; b) the best-ft straght lne by applyng the least square method; c) α

184 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION and β by means of (5A.3) and (5A.4). Accordng to the accuracy specfcatons formerly reported, the tme jtter on the samplng perod s some ppm of the perod tself and, hence, neglgble. In the smulatons, the effects of the remanng uncertanty sources have been assumed to be ndependent random varables wth unform dstrbuton and zero expected value. They are denoted by o n, o out, r n, r out, t d, hence, o O, O ], o O, O ], r R, R ], r R, R ], t [ u( T ) 3, u( T ) 3]. out [ out out n [ n n out [ out out d d d n [ n n The tme delay T d s constant n each tral. Indeed, td belongs to a set of totally correlated random varables. The same condton s assumed for o n, o out, r n, r out, gven that the offset s a constant contrbuton to all the samples of the consdered data set whereas the percentage of readng leads to a contrbuton proportonal to the sample value, wth constant proportonalty factor. The values of the above random varables are constant n a sngle tral but randomly vary over the whole range of test. The equpment descrbed has been used for the metrologcal characterzaton of a statstcal sample of nne voltage-to-voltage transducers LEM CV In the case of DC tests, nput sgnals varyng from 50 V to 750 V by steps of 50 V have been used. For each magntude value, the averagng procedure recalled n the prevous Secton has been appled to 200 samples acqured wth 2000 Sa/s frequency. As for AC tests, the RMS value of the snusodal nput was kept constant, equal to 500 V, whereas the sgnal frequency has been vared n the nterval [50 Hz, 3 khz]. The maxmum frequency has been chosen accordng to the requrements of both the IEEE Standard [12] and the European Standard [13] relevant to the harmoncs evaluaton. In each test, the above averagng procedure has been appled to fve perods of the nput sgnal. Table 5-7 a) and b) reports the maxmum absolute values A and B of α and β vs. frequency, along wth the relevant coverage ntervals assocated to the 95% coverage probablty. The confdence ntervals reported n Table 5-7 are effects of the uncertanty sources located n the equpment hardware on the estmaton of the ndexes α and β. Table 5-7 Values of A and B and relevant confdence ntervals for dfferent nput frequences (RMS nput value: 500 V; coverage probablty 95%) 5-7 a) Frequency A confdence nterval DC [1.5, 2.1] Hz [0.9, 1.5] Hz [1.0, 1.7] Hz [8.7, 9.5] b) Frequency B confdence nterval DC [0.54, 1.0]

185 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION 50 Hz [2.2, 5.7] Hz [5.1, 8.5] Hz [18, 23] 10-3 Fgure 5-28 and Fgure 5-29 show the hstograms of the relatve occurrences of α and β, respectvely. The hstograms have been determned for the transducers correspondng to the data reported n the 50-Hz row of Table 5-7. Both frequency dstrbutons are trangular and are vald for all the characterzed transducers. Ths s due to the lmted number of random varables consdered (assumed to have unform densty functons), on the bass of the Central Lmt Theorem N of occurrences [p.u.] x Fgure 5-28 Hstogram of the relatve occurrences of α at 50 Hz (p.u.: per unt) 0.1 α N of occurrences [p.u.] x 10-3 Fgure 5-29 Hstogram of the relatve occurrences of β at 50 Hz (p.u.: per unt) β

186 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION The standard devatons σ α and σ β of the random varables α and β can be expressed as functons of ndependent random varables wth unform dstrbuton: 2 x y 2 2 ( Rn + Rout ) 2 σ α = Var{} α = R 2 n (5A.8) x σ β = Var{} β = 4( x j Rn ) + On (5A.9) y 3 FS Such expressons can be demonstrated as follows. Recallng (5A.3): α = ( g g n ) / g n, by assumng g n = 1 provdes: α = g 1. If g s evaluated through the least mean square method, then t s: α = x y 1 2 x (5A.10) For the sake of smplcty, only the random varables o n, o out, r n, r out are taken nto account for each DAQ. Under these assumptons, the followng expresson for α holds: α = [ x + ( rn x + on )] [ x y + + ( r [ x + ( rn x + on )] x ( r y y + o ) + out 2 y + o y ( r out ] ) 1 + o out out n n 2 x + 2 x ( rn x + on ) x ) 1 (5A.11). In (5A.11) the second-order contrbutons are neglected. By also assumng that the nput sgnal s alternatve, we get: α x y + r x y + r x y [ 1+ rn + rout ] out n 1 = x + 2rn x (1 + 2rn ) x (5A.12) can be rewrtten as follows: α = α = x y 1 (5A.12) [ 1 + rn + rout ] x y [ 1 + rn + rout 2rn ] x y [ 1 + rn + rout ] (1 + 2r x y n ) x y (1 2rn ) 1 == 1 = x (1 2r ) x x ( r + r ) n out x x x n y 1, (5A.13) Under the assumpton of ndependent random varables wth unform dstrbuton, the expresson of the standard devaton σ α of α s:

187 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION {} ( ) Var n out n R R R x y x + = α = σ α (5A.14) whch s dentcal to (5A.8) under the assumptons: R n = R out (gven that the two DAQs are of the same model), and: 1 2 x y x. Equaton (5A.8) allows to estmate the uncertanty affectng α on the bass of the nomnal accuracy specfcatons of the DAQs. As far as the standard devaton σ β of the random varable β s concerned, let us recall (5A.4): { } { } ] [ max ] [ ] [ max out n out n u n u g n u = β Δ. By takng nto account the uncertanty contrbutons that affect the generc samples x and y, t becomes: ( ) β = out FS out FS n n out n out out o y r y o x r x r r x y x x y x o y r y ) ( ) ( max 2 2 out FS out FS n n out out out o y r y o x y x x r x y x x r x y x o y r x x y x y max. (5A.15) The approxmatons x = y and 1 2 = x y x, lead to: = β out FS out FS n n out out o y r y o x r x r o x x y x y 2 2 max 2 out FS out FS n out n out o y r y o o r r x x x y x y = ) ( 2 max 2 (5A.16) The term r out (y FS + o out ) can reasonably be neglected, n fact one could easly show that t only provdes second-order contrbutons. Hence, the fnal expresson for β becomes: FS n out n out y o o r r x x x y x y + + β = ) ( 2 max 2. (5A.17)

188 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION The expresson of the standard devaton σ β of β can then be obtaned by consderng the value j of the ndex that provdes β x j ( Rout + Rn ) + Oout + On σ β = Var{} β =. (5A.18) 2 3y Fnally, f we take R n = R out and O n = O out, then (5A.18) becomes: FS σ β = Var{} β = 4( x j Rn ) + On, (5A.19) y 3 FS whch s dentcal to (5A.9). It depends on the value of x j. However, f x j = y FS s taken, then: σ β 3 ( σ β 2 2 4R n + O = max 2 n ), (5A.20) 3y 2 2 FS Also the maxmum value of σ β only depends on the accuracy specfcatons of the samplng devces. Actually, the effect of the delay contrbuton u(t d ) has not been taken nto account n (5A.8) and (5A.9); anyway, by assumng x j = y FS, σ β has been overestmated. The results of (5A.8) and (5A.9) are compared n Table 5-8 wth the samplng standard devatons s α and s β of the dstrbutons n Fgure 5-28 and Fgure 5-29, respectvely. The comparson shows that (5A.8) and (5A.9) are useful to foresee the uncertanty affectng α and β wthout heavy computatons. Hence, an equpment mplementng the method descrbed can be properly desgned. Moreover, forecastng the values of s α and s β provdes nformaton on the mnmum values of α and β that the equpment can measure. Fnally, Table 5-8 also confrms that u(t d ) at 50 Hz s neglgble. Table 5-8 Comparson between s α, s β, and σ α, σ β of the random varables n Fgure 5-28 and n Fgure 5-29 s α σ α s β σ β

189 5. THE DISTRIBUTED MEASUREMENT SYSTEM FOR TRANSIENTS DETECTION References: [1] Model VD305A capactve voltage dvder, 2006 Pearson Electroncs, Inc. [2] J. Czajewsk, The Accuracy of the Global Postonng Systems, IEEE Instrumentaton and Measurement Magazne, March 2004, pp [3] ISO Gude to the expresson of uncertanty n Measurement, Internatonal Standardzaton Organzaton, Geneva (Swtzerland), [4] Evaluaton of Measurement data - Supplement 1 to the Gude to the Expresson of Uncertanty n Measurement propagaton of dstrbuton usng a Monte Carlo method, Jont Commttee for Gudes n Metrology, Fnal Draft September [5] Jaynes, E. T. Informaton theory and statstcal mechancs. Phys. Rev 106 (1957), [6] U. Poglano, Use of ntegratve analog-to-dgtal converters for hgh precson measurement of electrcal power, IEEE Trans. on Instr. Meas., vol. 50, no. 5, pp , Oct [7] M. Kampk, H. Laz and M. Klonz, Comparson of three accurate methods to measure AC voltage at low frequences, IEEE Trans. on Instr. Meas., vol. 49, no. 2, pp , Apr [8] IEC Std , Instrument transformers Part 1: Current transformers, Internatonal Electrotechncal Commsson, Geneva (Swtzerland), [9] IEC Std , Instrument transformers Part 2: Inductve voltage transformers, Internatonal Electrotechncal Commsson, Geneva (Swtzerland), [10] IEC Std , Instrument transformers Part 7: Electronc voltage transformers, Internatonal Electrotechncal Commsson, Geneva (Swtzerland), [11] IEEE Std , IEEE Standard for termnology and test methods for analog-to-dgtal converters, The IEEE, New York (USA), [12] IEEE Std , Recommended practce for montorng electrc power qualty, The IEEE, Pscataway (USA), November [13] EN 50160, Voltage characterstcs of electrcty suppled by publc dstrbuton systems, CENELEC, Bruxelles (Belgum), [14] A. Ferrero, M. Lazzaron, S. Salcone, A calbraton procedure for a dgtal nstrument for power qualty measurement, IEEE Trans. on Instr. Meas., vol. 51, no. 4, pp , August [15] N. Locc, C. Muscas, E. Ghan, Evaluaton of uncertanty n dgtal processng of quantzed data, Measurement, vol. 32, no. 4, December 2002, pp [16] A. Ferrero, S. Salcone, An nnovatve approach to the determnaton of uncertanty n measurement based on fuzzy varables, IEEE Trans. Instr. Meas., vol.52, no. 4, pp , [17] Natonal Instruments, The measurement and automaton catalog, 2004, Austn (USA), [18] 3GPP specfcaton Multplexng and Multple Access on the Rado path

190

191 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER 6. Desgn ad Characterzaton of an Electrc feld based Medum-Voltage Transducer Power qualty measurement nstruments, especally those n portable packages, are generally provded wth nputs desgned for low-voltage applcatons. Some permanently nstalled PQ measurement nstruments are mounted at a dstance from the pont of the crcut where the parameters are to be measured. In both cases, a sutable transducer mght be needed, to step down the voltage, to solate the nput crcuts from the system voltage, or to transmt the sgnals over some dstance. To accomplsh any of these functons, a transducer may be used, provded that ts characterstcs are sutable for the parameter of nterest. In low-voltage systems, PQ measurement nstruments are generally connected drect to the voltage pont of nterest, but transducers are often used for current measurements. In medum- and hgh-voltage systems, transducers are used for both voltage and current PQ measurements. There are two mportant concerns usng transducers: sgnal levels: sgnals levels should use the full scale of the nstrument wthout dstortng or clppng the desred sgnal; frequency and phase response: these characterstcs are partcularly mportant for transent and harmonc measurements. In order to avod ncorrect measurements the full-scale ratng, lnearty, frequency and phase response, and burden characterstcs of the transducer should be carefully consdered [1] Voltage transducers The most common voltage transducer s the voltage transformer. Two types of voltage transformers can be consdered: those used by protectve relay crcuts, and those used by meterng crcuts. The frst type s szed so as to provde a correct response even n the case of overvoltages due to an unbalanced short crcut. The second, n contrast, s desgned to protect meters from network overvoltages. In the latter category, n case of saturaton, dstorton of the delvered sgnal wll occur. Where montorng s attached to a voltage transformer whch s also used for other functons (for example, meterng), one must be careful that the addtonal burden do not affect the calbraton or uncertanty of such other functons. One should be careful when makng connectons to the secondary crcut of a transformer used for a protectve relay. Connecton errors mght cause the relay to nadvertently trp. In general, transformer-type electromagnetc voltage transducers have frequency and transent responses sutable up to typcally 1 khz; but the frequency range may

192 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER sometmes be lmted to well below 1 khz, and sometmes may extend to a few klohertz. Smple capactor dvders can have frequency and phase responses that are sutable up to hundreds of klohertz or even hgher; however, n many applcatons a resonant crcut s ntentonally added, makng the frequency response of the capactve dvder unsutable for measurements at any frequency other than the fundamental. Resstve voltage dvders may have frequency and phase response sutable up to hundreds of klohertz. However, they may ntroduce other problems, for example, the capactve load of the measurement nstrument can nfluence the frequency and phase response of the resstve voltage dvders. There are two mportant concerns that must be addressed when selectng transducers for a.c. mans transents. Frst, sgnal levels should use the full scale of the nstrument wthout dstortng or clppng the desred sgnal. Second, the frequency response (both ampltude and phase) of the transducer should be adequate for the expected sgnal. Voltage transducers should be szed to prevent measured dsturbances from nducng saturaton. For low-frequency transents, ths requres that the knee pont of the transducer saturaton curve be at least 200 % of the nomnal system voltage. The frequency response of a standard meterng class Voltage transducer depends on ts type and the burden appled. Wth a hgh mpedance burden, the response s usually adequate to at least 2 khz, but t can be less. Capactve-coupled voltage transformers generally do not provde accurate representaton of any hgher frequency components. Hgh-frequency transent measurements requre a capactor dvder or pure resstve dvder. Specal purpose capactor dvders can be obtaned for measurements requrng accurate characterzaton of transents up to at least 1 MHz. For both voltage and current, the spectra of common test waveforms for a.c. mans transents (see Fgure 6-1) contan frequences that range up to approxmately 10 MHz (lastng for 200 µs), wth large ampltudes up to 1 MHz (lastng for 2 ms). For end-use a.c. mans connectons, the ampltudes of common test waveforms range up to 6 kv, and up to 5 ka. The samplng rate must therefore be at least twce the maxmum frequency of the waveform; also, the correspondng ant-alas flter must have approprate characterstcs. On the one hand, for low voltage power systems there are dfferent solutons to measure power qualty, n fact the market offers several commercal transducers featurng wde bandwdth and lnear characterstc for reasonable costs. Most of the transducers nclude an electronc crcut, hence they can be consdered actve devces: current transformers explot Hall-effect based sensors or Rogowsk col followed by the analog condtonng crcut, whereas actve voltage transformers can be made of resstve dvders combned wth sutable solatng amplfers or use the same Hall-effect probe wth a shunt resstor at

193 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER the nput of the condtonng stadum. Passve current transformers mprovng performance respect to the tradtonal ones are wndngs wth compensatng capactors. Fgure 6-1 Frequency spectrum of typcal representatve transent test waveforms On the other hand, as a consequence of the recent recommendatons of natonal regulators [2, 3], some utltes have shown ncreasng nterest n power qualty measurements, wth reference to transmsson and especally dstrbuton lnes. Presently, Italy, Norway, Portugal and Slovena are usng so-called voltage qualty montorng system on both dstrbuton and transmsson lnes. In Hungary only dstrbuton lnes are montored whereas Czech Republc and Span are plannng measurement actvtes. Unfortunately, wdeband commercal transducers sutable for such applcatons are expensve and hard to nstall n usual medum-voltage substatons. Therefore, measurement campagns are often lmted to a very small porton of network. As far as the scentfc communty proposals are concerned, the use of optcal voltage transducers s dealt wth n some papers [4-6]. An nterestng soluton to optcally supply actve voltage transducers s proposed n [7]. In order to decrease sgnfcantly the cost of the dstrbuted measurement system developed durng the PhD research actvty, a part from reducng the number of remote unts to be nstalled on the network by adoptng the wavelet-based algorthm, t s mportant to fnd the cheapest verson of the hardware needed n each channel of a slave staton. The most expensve block n the measurement chan descrbed n cap 5a s the voltage transducer,.e. the capactve voltage dvder by Pearson Electroncs. The research has then been focused to develop a dfferent transducer whch could perform smlarly but cost much less than t. The man requrements to be fulflled were: a)

194 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER bandwdth sutable also for transent measurements; b) good accuracy; c) solaton level sutable for MV systems; d) easy and fast nstallaton on the lne to be montored. The soluton adopted s an actve voltage transducer obtaned as combnaton of an electrc feld probe and the sutable condtonng crcut, so that no electrc contact s needed between the devce nput and the conductor of the lne on whch t s nstalled. In ths case the IEC Standard [8] relevant to Electronc Voltage Transformers has been taken nto account as reference document for: ratngs settng (standard values of prmary voltage and output voltage); accuracy class desgnaton (accordng to the voltage error and phase dsplacement ntroduced by the nstrument); nsulaton requrements; and the correspondng tests Electrc feld strength meters The purpose of IEEE Standard [9] s to establsh unform procedures for the measurement of power frequency electrc and magnetc felds from alternatng current overhead power lnes and for the calbraton of the meters used n these measurements. A unform procedure s a prerequste to comparsons of electrc and magnetc felds of varous ac overhead power lnes. These procedures apply to the measurement of electrc and magnetc felds close to ground level, but even to electrc feld measurements near an energzed conductor or structure. Electrc feld strength meters consst of two parts: the probe or feld sensng element and the detector whch processes the sgnal from the probe and ndcates the RMS value of the electrc feld strength n unts of V/m usng an analog or dgtal dsplay. For commercally avalable free-body meters, the detector s usually contaned n, or s an ntegral part of, the probe. The probe and detector are ntroduced nto an electrc feld on an nsulatng handle. The detector measures the steady-state nduced current or charge oscllatng between the conductng halves (electrodes) of the probe. In Standard IEEE [10] the followng three types of electrc feld meters are consdered: a) The free-body meter b) The ground reference meter c) The electro-optc meter. When measurements of the electrc feld strength are performed, the observer must be suffcently removed from the probe to avod sgnfcant perturbaton of the feld at the locaton of the probe. Free-body and electro-optc type meters should be suffcently small so that the sze of the probe does not sgnfcantly perturb the charge dstrbutons on boundary surfaces generatng the electrc feld,.e., energzed and grounded surfaces. Although feld meters are calbrated n nearly unform electrc felds, the feld that s measured need not be very unform. Electrc feld meters measure the projecton of the

195 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER oscllatng (lnearly polarzed) or rotatng (ellptcally or crcularly polarzed) electrc feld vector onto the electrcal axs of the probe (the axs of greatest electrc feld senstvty). The three types of meters used to measure the electrc feld strength from ac power lnes are descrbed n the followng Free-Body meter It measures the steady-state nduced current or charge oscllatng between two halves of an solated conductve body n an electrc feld. The free-body meter s sutable for surveytype measurements because t s portable, allows measurements above the ground plane, and does not requre a known ground reference. Therefore, ths type of meter s recommended for outdoor measurements near power lnes. The sze of the probe should be such that charge dstrbutons on the boundary surfaces generatng the electrc feld (energzed and ground surfaces) are, at most, weakly perturbed when the probe s ntroduced for measurement. The electrc feld should be approxmately unform n the regon where the probe wll be ntroduced. Probes can be of any shape; however, meters commercally avalable are generally n the shape of rectangular boxes, wth sde dmensons rangng from ~7 to ~20 cm, as represented n Fgure 6-2. The meters are calbrated to read the RMS value of the power frequency electrc feld component along the electrcal axs (the axs of greatest electrc feld strength senstvty). There also exst free-body meters desgned for remote dsplay of the electrc feld strength. In ths case, a porton of the sgnal processng crcut s contaned n the probe and the remander of the detector s n a separate enclosure wth a dgtal dsplay. A optcfbre lnk connects the probe to the dsplay unt. Ths type of probe s also ntroduced nto an electrc feld on an nsulatng handle. Fgure 6-2 Geometres of E-feld probes: (a) sphercal probe (b) commercal probes In order to characterze the nstrumentaton adequately, the manufacturer should provde a detaled descrpton of the electroncs, as well as other relevant nformaton. For example, f the feld meter readng has a temperature dependence, the temperature coeffcent should be provded. Ths permts the operator to correct E-feld readngs made outdoors usng an nstrument calbrated at room temperature. If the electrcal axs of the

196 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER feld strength meter s not concdent wth the geometrc axs, the departure n degrees and drecton shall be specfed. Brefly, the theory of operaton of free-body meters can be understood by consderng an uncharged conductng free body wth two separate halves ntroduced nto a unform feld E. The charge nduced on one of the halves s: r r Q = D da S / 2 (6.1) Where D r s the electrc dsplacement and da r s an area element on half of the body wth total surface area S. The case of sphercal geometry (Fgure 6-2) yelds the result 2 Q = 3πa ε0e where a s the radus of the sphere and ε 0 s the permttvty of free space. (6.2) The surface charge densty s gven by 3ε 0 Ecosθ. Integraton over the hemsphere gves the prevous equaton. For less symmetrc geometres, the result can be expressed as: Q = kε 0E (6.3) where k s a constant dependent on geometry. Sensng electrodes resemblng cubes and parallel plates (Fgure 6-2) have been employed. If the electrc feld strength has a snusodal dependence, for example, E 0 snωt, the charge oscllates between the two halves and the current s gven by: dq I = = kωε 0 E0 cosωt (6.4) dt If there are harmoncs n the electrc feld, there wll be an addtonal term on the rght sde of equaton (6.4) for each harmonc. Because of the dfferentaton operaton n equaton (6.4), each of the addtonal terms wll be weghted by the assocated harmonc number. As for the magnetc feld meter case, t s necessary for the detector to perform the nverse mathematcal operaton, namely ntegraton, to recover the electrc feld waveform. Ths s accomplshed by ntroducng a stage of ntegraton. For example, an ntegratng amplfer or a passve ntegratng crcut combned wth a voltmeter could be used as a detector. The frequency response of the probe-ntegratng detector combnaton should be made flat over the frequency range of nterest. Flters should be used to exclude sgnals outsde of the frequency range of nterest. It should be noted that the unform E-feld drecton serves as an algnment axs for the feld probe and that durng feld measurements ths axs should be algned wth the feld component of nterest. The constant k can be thought of as a feld strength meter constant and s determned by calbraton. For more exact results, a second term not shown should be added to the rght-hand sde of equaton (6.4) because of the presence of the delectrc handle held by the observer. The nfluence of the handle, representng a leakage

197 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER mpedance, and the perturbaton ntroduced by the observer are taken to be neglgble n the above dscusson. The detector, although calbrated to ndcate the RMS value of the power frequency feld, may, dependng on the detector crcut desgn, measure a) a quantty proportonal to the average value of the rectfed power frequency sgnal from the probe; b) The true RMS value of the sgnal. The response of the detector to harmonc components n the E-feld also depends on the desgn of the detector crcut. For example, n case a), because of the sgnal-averagng feature, an analog dsplay wll not necessarly ndcate the RMS value of the composte E- feld waveform (fundamental plus harmoncs). For case b), the true RMS value of the electrc feld strength wth harmoncs could be observed f the detector crcut contaned a stage of ntegraton. The frequency response of the free-body meter can be determned expermentally by njectng a known alternatng current at varous frequences and observng the response. The rated accuracy of the detector at power frequency s a functon of the stablty of ts components at a gven temperature and humdty and s generally hgh (<0.5% uncertanty) Ground-Reference-Type meter Such a devce measures the current-to-ground from a flat probe ntroduced nto an electrc feld. Flat ground-reference-type meters can be used only under specal condtons. Electrc feld strength meters ntended for characterzaton of rado-frequency electrc felds should not be used to measure the electrc feld strength from ac power lnes. Fgure 6-3 Two desgns for flat probes used wth ground-referenced electrc feld meters

198 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER Ground reference meters determne the electrc feld strength by measurng the current or charge on the sensng surface of a flat probe. Such meters are normally used to measure the electrc feld at ground level or on flat conductng surfaces that are at ground potental. Two probe desgns, drawn n Fgure 6-3 have been employed. One desgn makes use of a sngle flat conductor wth an solated central secton that serves as the sensng surface. Small versons of ths type of probe have been made wth double-clad prnted crcut board. A second desgn conssts of two parallel plates separated by a thn sheet of nsulaton, wth the top plate actng as the sensng surface. From Gauss Law, the charge, Q, nduced on a sensng surface wth area A, s: Q = ε 0 EA (6.5) where E s the average electrc feld strength across the sensng surface; ε 0 s the permttvty of free space. Assumng that E vares snusodally wth angular frequency ω,.e., E = E o snωt, the resultng nduced current s gven by: I = ωε 0 EAcosωt (6.6) Electro-optc meter The probe used n ths case s subject to the Pockels effect when t s ntroduced nto the electrc feld to be measured. Electro-optc feld meter s smlar to the free-body meter n that t s sutable for survey type measurements, allows measurements at most ponts above the ground plane, and does not requre a ground reference potental. The probe, whch s separate from the detector, can be supported n the feld wth an nsulatng handle. The probe and detector are connected wth optcal fbers through whch lght from the detector s routed to and from the probe. In general, the probes are small n dmenson (~2 cm) compared to free-body meter probes and ths permts measurements closer to conductng surfaces because of the smaller nteractons wth the surface charge dstrbutons. However, whle smaller n sze, Pockels effect probes have less senstvty to electrc felds (~5 kv/m and hgher) compared to freebody meters (~1 V/m and hgher) and are more expensve to fabrcate

199 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER Fgure 6-4 Probe for Pockels-effect electrc feld meter - The ampltude of the modulaton as lght passes through the Pockels crystal and other optcal elements provdes a measure of the electrc feld E. Fgure 6-4 shows a sketch of a Pockels effect probe and ts consttuent components. Lght orgnatng n the detector s sent to and from the probe va optcal fbers. The electrc feld E nduces a brefrngence n an approprately orented delectrc (Pockels) crystal that causes the ntensty of the lnearly polarzed lght to be modulated accordng to the relaton where: I I t [ 1+ sn( E' )] = F (6.7) 2 I t s the transmtted lght I s the ncdent lght E s the electrc feld n the crystal F s equal to λ / 2πn 3 c e l; λ s the wavelength of lght; n s the ndex of refracton; l s the crystal thckness; c e s the electro-optc coeffcent of the crystal. For equaton (6.7) to hold, t s assumed that the crystal has no ntrnsc optcal actvty. Equaton (6.7) shows that the ampltude of lght modulaton s a functon of the electrc feld n the crystal that, n turn, s dependent on the external feld E. Because the lght transmsson tracks the waveform of the electrc feld, a stage of ntegraton s unnecessary n the detector to approprately process sgnals due to harmoncs that may be n the electrc feld. The Pockels crystal s sometmes coated wth transparent electrodes to permt measurements of voltage usng the Pockels effect. Electro-optc meters may be battery or mans operated

200 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER Calbraton of electrc feld strength meters Parallel plate structures, sngle ground plates wth guard rngs, and current njecton crcuts have all been used for calbraton purposes. Each s now brefly descrbed. Parallel plates Unform feld regons of known magntude and drecton can be created for calbraton purposes wth parallel plates, provded that the spacng of the plates, relatve to the plate dmensons, s suffcently small. The unform feld value E 0 s gven by V/t, where V s the appled potental dfference and t s the plate spacng. As a gude for determnng plate spacng, the magntudes of the electrc feld strength E, normalzed by the unform feld, that s E/E 0, at the plate surface and mdway between sem-nfnte parallel plates are plotted as a functon of normalzed dstance x/t from the plate edge n Fgure 6-5. Because nearby ground surfaces are always present, gradng rngs have been employed to grade the feld at the permeter of the structure and to provde solaton from surroundng perturbatons. No exact theoretcal treatment of the problem s avalable for rectangular geometres, but analytcal solutons do exst for structures of cylndrcal symmetry. Fgure 6-5 Calculated normalzed electrc feld at plate surfaces and mdway between plates as functon of normalzed dstance from edge of plate Parallel plate structures can be energzed wth one plate at zero potental or both plates can be energzed usng a center tapped transformer. For example, stretched metal screens on 3 m x 3 m frames wth a 1 m separaton and four gradng rngs have been used to form a parallel plate structure. Potentals are appled to the gradng rngs usng a resstve dvder. Resstors that effectvely short out stray capactance between the gradng rngs and nearby surfaces are used. Theoretcal consderatons and expermental measurements ndcate that energzaton of the plates usng a center tapped transformer

201 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER provdes a feld that s more mmune from nearby sources of perturbaton than other energzaton schemes. The electrc feld strength meter shall be calbrated perodcally, wth the frequency of calbratons dependng n part on the stablty of the meter. The meter shall be placed n the center of a parallel plate structure wth the nsulatng handle normally used durng measurements. The dmensons of the structure should be 1.5 m x 1.5 m x 0.75 m spacng. Wth these dmensons, no gradng rngs (or resstor dvders) are necessary to obtan a calbraton feld that s wthn 1% of the unform feld value V/t. It s assumed that the largest dagonal dmenson of the electrc feld strength meter to be calbrated s no larger than 23 cm. The dstance to nearby ground planes (walls, floors, etc.) shall be at least 0.5 m. The dmensons of the calbraton apparatus may be scaled upward or downward for calbraton of larger or smaller electrc feld strength meters Man sources of measurement uncertanty The measurement uncertanty durng practcal outdoor measurements usng commercally avalable free-body meters s typcally near 10%, although ths fgure can be reduced under more controlled condtons. The most lkely sources of major errors are dffculty n postonng the meter, readng errors, handle leakage n some cases, temperature effects, and observer proxmty effects. Because of nteractons that can occur between the feld meter probe and surface charge dstrbutons on nearby conductng surfaces, the electrc feld measurement can be sgnfcantly perturbed f the probe s brought too close to the surface. Calculatons show that ths perturbaton for a sphercal probe s reduced to near 0.5% when the dstance between a ground plane and the probe centre s three probe rad. Therefore, a sphercal probe s not expected to have sgnfcant measurement error f a dstance of two probe dameters s mantaned between the probe and conductng surfaces. The dameter of probes wth rectangular geometres can be conservatvely estmated as the largest dagonal dmenson of the probe. Asymmetres n the desgn of an electrc feld meter probe can change the drecton of the electrcal axs (axs wth greatest electrcal senstvty) wth respect to the geometrcal axs. Measurements performed wth such an nstrument may be more or less mmune to observer proxmty effects. In such cases, the observer proxmty effect should be quantfed before the feld meter s used for measurements. Because magnetc felds are typcally produced at the same tme as electrc felds, electrc feld meters should be desgned so that they are not sgnfcantly affected by magnetc felds at the levels antcpated n a gven measurement envronment. The col system descrbed n standard IEEE 1308 [10] for producng magnetc felds can be used to check for mmunty to magnetc felds

202 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER Electrcal leakage va a grounded observer and surface contamnaton on the nsulatng handle of the meter may perturb the electrc feld beyond the normal geometrc perturbaton produced by the electrcally floatng probe. To check for handle leakage, the electrc feld meter should be orented wth ts axs perpendcular to the drecton of a known feld. Sgnfcant electrcal leakage would cause a nonzero readng. Such a readng, expressed as a percent of the actual feld, would represent the order of magntude of the uncertanty that could be caused by ths mechansm. It s assumed for ths check that the electrcal and geometrcal axes concde. Under hgh humdty condtons, a layer of surface condensaton may form on parts of an electrc feld meter. The major source of uncertanty comes from handle leakage through the mountng nsulaton to one of the electrodes. If sgnfcant, ths leakage wll greatly ncrease the currents nduced n the probe and the resultng feld meter readng. A much smaller uncertanty s assocated wth leakage between the two sensng electrodes, whch would reduce the readng of the feld meter. The feld meter, ts handle assembly, and ts nternal nsulaton should be kept clean and dry to mnmze errors due to leakage currents. The nfluence of ambent humdty on the performance of feld meters can be determned by applyng the current njecton technque (free-body meters) or voltage njecton technque (electro-optc meters) wth the feld meter n an envronmental chamber. The dependence on humdty can be determned by montorng the feld meter response as a functon of humdty whle holdng the njected current (voltage) constant. Envronmental chamber tests of free-body meters wth analogue dsplays have shown that the feld meter readng can change by as much as 8% over the temperature range 0 C to 40 C. As for the magnetc feld meter case, f extreme dfferences n temperature are antcpated at a measurement ste compared to the temperature at the tme of calbraton, the effects of temperature should be known or may need to be quantfed. In order to determne the total uncertanty assocated wth RMS measurements of the electrc feld strength n dfferent measurement envronments, there should be an approprate accountng of the varous sources of uncertanty. Possble sources of uncertanty have been dentfed above. Many sources of uncertanty can be made neglgble or, dependng on the type of meter, may not apply n a gven measurement stuaton. In any event, the combned uncertanty of the sgnfcant sources of uncertanty should be taken as the square root of the sum of the squares

203 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER 6.3. The voltage transducer Operatng prncple The voltage transducer reles on the measurement of the electrc feld rradated by the consdered conductor. To do ths, a devce based on the schematc block dagram n Fgure 6-6 was mplemented. Electrc feld sensor Condtonng crcut output voltage DC power supply Voltage transducer Fgure 6-6 Schematc block dagram of the voltage transducer It conssts n three man elements: a) an electrc feld probe; b) a condtonng crcut; c) a dc bpolar power source. Accordng to what reported n prevous Secton, the electrc feld sensor can be consdered as a Ground-reference and un-sotropc type. The probe s a metallc surface normal to the ncdent electrc feld component and connected to ground through a proper shunt resstance. A current (t) flows from the plate dranng to ground the free charges cumulated on ts surface. As recalled n prevous secton, n other words (t) s proportonal to the frst dervatve of the ncdent electrc feld. Moreover, f ) the dstance between the conductor and the probe s constant, ) the delectrc mean nterposed between them s fxed and ) the nfluence of external electrc feld sources can be neglected, the current (t) can be assumed also proportonal to the frst dervatve of the voltage u(t) to be measured,.e. to the source of the electrc feld: du( t) ( t) (6.8) dt Therefore, the condtonng crcut ntegrates (t) provdng an output voltage proportonal to u(t). Sutable mpedance matchng and sgnal amplfcaton are also operated Prototype realzaton To fulfl (6.8), three ssues must be tackled n desgnng the transducer prototype: ) ensure the recommended electrcal nsulaton between the transducer and the voltage source; ) sheld the devce from other electrc feld sources; ) keep constant the dstance between the conductor and the sensor

204 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER As for ) the soluton adopted conssts n placng the voltage transducer nto an nsulatorlke case so that t s nstalled drectly on the conductor (Fgure 6-7). Ths way, also problem ) s solved gven that the dstance between the probe and the source s fxed. Moreover, n ths condtons also the delectrc nterposed between the electrc feld probe and source s known and stable as far as electrcal and envronmental nfluences are concerned. Fgure 6-7 The nsulator-lke case contanng the electrc-feld based transducer The same could not be sad wth ar nterposed between them, n fact n ths case temperature, drt and humdty, just to gve an example of the man envronmental factors, can affect strongly the performance of the sensor. Fgure 6-8 Secton of the voltage transducer: the shelded probe and crcut are covered by the nsulator-shaped resn. The nsulator was szed for a nomnal RMS voltage of 17.5 kv, whch s a typcal value n Italan dstrbuton networks. In accordance wth [8], the correspondng rated nsulaton level requres to pass tests wth a rated lghtnng-mpulse of 95 kv and a power-frequency wthstand voltage of 38 kv for 60 s. As for ), both the sensor and the condtonng crcut were shelded by a grounded metallc case, ncluded nsde the nsulator-lke one. The sheldng box has been provded

205 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER wth a sutable aperture facng the MV copper bar n order to let only the electrc strength lnes generated by the source of nterest enter the sheld. As far as the condtonng crcut s concerned, the easy prncple of operaton does not turn nto an mmedate correct realzaton of t. Frst of all, the nput buffer has to be chosen wth very hgh nput mpedance, so that the probe s not loaded and also the value of the shunt resstance n parallel to the nput pns s not sgnfcantly modfed. The gan of the frst stadum does not need to be hgher than 10, and ths mproves the stablty of the crcut. Snce the voltage measured as nput s proportonal to the frst-order tme dervatve of the prmary voltage, the equvalent model of the sequence electrc feld probe - shunt resstor corresponds to an deal dervatve-operator. Desptes, the analog crcut used to obtan the output voltage cannot be an deal ntegrator, because any bas currents and offset voltage of the operatonal amplfer used n the ntegrator or non-whte nose superposed to the nput would lead the output to saturaton. In fact, the ntegrator gan ncreases as frequency decreases, thereby amplfyng the low-frequency random nose and zero-frequency offset drft. The ntegrator gan needs to be reduced for frequences below whch the measurement accuracy s not sgnfcantly affected. A large resstor R 2 s put across the capactor C 1 n order to provde DC feedback for stable basng. The effect of such confguraton s to roll off the ntegrator acton at very low frequences (f < 1/R 2 C 1 ). The transfer functon of the ntegrator shown n Fgure 6-9 s gven by the followng expresson: V V out n R2 1 = (6.9) R (1 + jωr C ) The ntegrator desgn,.e. the choce of sutable values for R 1, R 2 and C 1, s actually a trade off between gan and ntegratng pole poston. The accuracy of the ntegraton s strongly affected by the value of the ntegrator pole, placed at f=1/2πr 1 C 1. In ths applcaton the pole corresponds to a nomnal frequency of 50 mhz, so that the performance of the transducer are not degraded: the phase dsplacement ntroduced by the crcut at 50 Hz s lower than 3 mrad. As above explaned, an operatonal amplfer confgured to work as ntegrator greatly attenuates the nput AC sgnal. To compensate such strong reducton of the component of nterest, the DC gan of the amplfer,.e. A=R 2 /R 1, s set equal to 100. A hgher value would gve rse to wrong results because both the stablty of the stadum would not be assessed and the bas components would be too much amplfed respect to the AC nput sgnal

206 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER R 2 C 1 R 1 - v n + v out Fgure 6-9 Operatonal amplfer confgured as ntegrator Fnally, the last amplfer has been used to regulate the overall gan of the condtonng crcut, obtanng a standard rated secondary voltage equal to 1.625/ 3 V [10],.e. when the prmary phase-to-ground voltage s 17.5/ 3 kv (phase-to-phase voltage equal to 17.5 kv) the output s / 3 V. Fgure 6-10 Drecton of the electrc axs of the feld probe, concdent wth the geometrcal axs of the transducer Expermental setup The expermental campagn was programmed to nvestgate on four man characterstcs of the devce performance: a) voltage error; b) phase dsplacement; c) bandwdth; d) mmunty to other voltage sources present n any three-phase MV/lv cabn. The relevant setups are descrbed n the followng

207 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER Voltage error MV Conductor HI TUT Inductve Voltage Transformer Reference Capactve Dvder Varable Voltage Supply DMM LO out gnd out gnd DMM DC power + gnd supply - Fgure 6-11 Schematc representaton of the crcut for the voltage error evaluaton The crcut shown n Fgure 6-11 was mplemented, where TUT denotes the voltage transducer under test and DMM an HP 3458A dgtal multmeter. The DMM nomnal accuracy specfcatons n the consdered range U FS =10 V and for f=50 Hz, are α=0.09% of readng plus β=0.06% of range [11]. The low-voltage sde of a standard nductve voltage transformer (70 kv/100 V) s suppled wth a snusodal 50-Hz voltage waveform. The transformer output (hgh-voltage sde) fed both a plan copper bar, where the TUT s nstalled, and a reference capactve dvder made by Pearson Electroncs,.e. the transducer that should be substtuted by the one under test. Its specfcatons (n ol and nto 1MΩ load) are: nomnal rato = 5000:1, bandwdth = 30 Hz 4 MHz, droop rate = 0.02% / µs, usable rse-tme = 100 ns. The outputs of both the capactve dvder and the TUT were measured by the DMMs, controlled by a personal computer va IEEE 488 nterface. The control program confgured each DMM to carry out 100 measurements of RMS for every value of the test voltage (U test ); n order to mnmze the random uncertanty contrbuton of the test bank, the DMM gves as output the mean value of 10 subsequent RMS readngs. Table 6-1 Measurements of prmary and secondary voltages U test [kv] U n [kv] u(u n ) [kv] U out [V] u(u out ) [V]

208 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER In accordance wth Fgure 6-11, one hundred measurements of both the TUT and the capactve dvder output were taken for each testng voltage. It must be noted that U n refers to a phase-to-ground voltage. As a consequence, 10 kv corresponds at a phase-tophase voltage of about 17.5 kv, whch s the rated prmary voltage of our devce. Table 6-1 shows the obtaned results, where u( ) denotes the standard uncertanty affectng ( ), evaluated accordng to [13] by takng nto account both the nstrument accuracy and the standard devaton of the 100 measurements. On the bass of the tests results reported n Table 6-1, the transformaton rato K of the transducer to be characterzed s computed accordng to the followng expresson: K U out out n ref = = (6.10) U n U U K ref where U n, out are the RMS of the voltage sgnals at the nput and output of the TUT, whereas U ref s the RMS voltage measured at the output of the capactve dvder and K nref ts nomnal rato. The results obtaned for each U test feature very smlar frequency dstrbutons; for example Fgure 6-12 reports the 20 classes hstogram relevant to U out measured for U test = U n = 10kV. Accordng to [8] the voltage error of the transducer under test, expressed n percent, s defned as: ε U = K n U U out n U n 100 (6.11) number of occurences U out [V] Fgure 6-12 Frequency dstrbuton of the secondary U out measured for U test = U n

209 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER As t can be drawn from Fgure 6-12 the man contrbuton to u(u out ) s due to the expermental standard devaton,.e. around 80% of the standard uncertanty. Smlar consderatons hold for all the prmary voltages and also for u(u n ). Table 6-2 Evaluaton of the Voltage Error ε u for the dfferent test voltages. U test [kv] ε u [%] u(ε u ) [%] Table 6-2 shows the voltage error values for dfferent prmary voltages along wth ther relevant standard uncertantes u(ε u ). As well known, ISO document [8] classfes voltage transducers accordng to ther accuracy evaluated on the bass of the voltage and phase error. The frst one s requred to be lower than a gven value (accordng to the accuracy class) for any voltage between 80% and 120% of the rated prmary voltage, that s between 8 kv and 12 kv n the case of the prototype under test. Results reported n Table 6-2 lead to conclude that, at least for the voltage error, the transducer can be consdered as belongng to class Phase error MV Conductor HI TUT Inductve Voltage Transformer Reference Capactve Dvder Varable Voltage Supply LO out 1 gnd out 2 DAQ board gnd DC power + gnd supply - Fgure 6-13 Schematc representaton of the crcut for the phase error evaluaton

210 6. DESIGN AND CHARACTERIZATION OF AN ELECTRIC FIELD BASED MV TRANSDUCER The crcut shown n Fgure 6-13 was mplemented, where DAQ board refers to a 16-bt dgtal acquston board featurng a smultaneous samplng. Ths way, no addtonal delay s added to that of the TUT. The outputs of both the capactve dvder and TUT were smultaneously acqured by the DAQ board. A samplng frequency of 200 ksa/s and an observaton nterval of 200 ms (40,000 samples) were used. The personal computer ran a smple algorthm to extract the 50-Hz components and to compute the phase error. The measurement procedure was carred out 1,000 tmes amng at takng nto account the effect of random contrbutons to the uncertanty. Fourer analyss was used to extract the 50-Hz components, thus avodng the contrbuton of eventual prmary voltage dstortons. The phase error φ u s gven by [8]: ϕ u = ϕ out ϕ n Table 6-3 Results of tests for the phase error evaluaton of the TUT U test [kv] φ u [mrad] u(φ u ) [mrad] (6.12) number of occurences Fgure 6-14 Frequency dstrbuton of the phase dsplacement between nput and output of the TUT obtaned n the case of U test = U n = 10 kv