Fault resistance sensitivity of sparse measurement based transmission line fault location

Transcription

1 ault resstance senstvty of sparse measurement based transmsson lne fault locaton Papya Dutta Dept. of Electrcal and Computer Engneerng Texas A & M Unversty College Staton, TX, USA papya8@tamu.edu Abstract Tradtonal transmsson lne fault locaton methods requre measurements from at least one end of the faulted lne. Measurements from all the ends of the faulted lne are desrable but not always avalable. Sparse measurement based fault locaton scheme usng phasor measurements from dfferent substatons located n the vcnty where the fault has occurred can be appled f the measurements are not avalable from any of the lne ends. ault resstance s one of the major sources of uncertanty n transmsson lne fault locaton estmaton. Ths paper presents a correcton scheme to reduce mpact of fault resstance on sparse measurement method. I. INTRODUCTION Transmsson lnes exposed to dfferent weather, as well as human and anmal contact are subject to several types of faults whch are caused by random and unpredctable events. Protectve relays placed at both (all) ends of the lne sense the fault mmedately and solate the faulted lne by openng the assocated crcut breaers. To restore servce after a fault, an accurate locaton of the fault s needed to help the mantenance crew fnd and repar the faulted lne secton as soon as possble. ault may occur between transmsson lne phases or have a ground return path. When a phase-to-phase fault occurs, the fault current flows through arc resstance and f the fault s a ground fault the current path ncludes earth resstance (conssts of tower resstance, tower footng resstance and ground return path) also. ault resstance s the combned resstance whch appears n the fault current path. Ths s an uncertan parameter as both the arc resstance and earth resstance depend on many parameters that are sometmes very hard to predct. Dstance relay algorthm selectvty may suffer from the combned effect of fault resstance and load current whch s nown as reactance effect [1]. Such algorthms assume that the fault current s n phase wth measured current. Presence of remote nfeed complcates the stuaton. Taag et al. [-3] decomposed the faulted networ to pre-fault and pure-fault networ and tae some assumptons to elmnate fault resstance part from the crcut equaton. Another one-end method usng quadratc formula to elmnate fault resstance s Mladen Kezunovc Dept. of Electrcal and Computer Engneerng Texas A & M Unversty College Staton, TX, USA ezunov@ece.tamu.edu ntroduced n [4] yeldng much more accurate result. Usng one-end data to estmate fault resstance by modelng the arc s dscussed n [5-6]. [7] s based on equalzng voltage of fault pont from both ends of the lne based on measurements from both ends and thus elmnates the mpact of fault resstance. A settngs free fault locaton method usng synchronzed samples from both ends of the lne s completely ndependent of fault resstance, whch s not used to develop the algorthm [8]. Typcally dgtal fault recorders (DRs) or dgtal protectve relays (DPRs) are placed n substatons and they record current, voltage and status sgnals on occurrence of an event le fault. Due to the lac of measurement transformers n certan transmsson lne confguratons such as tapped lnes avalablty of measurements from at least one end of the lne becomes a problem. If the measurements from other ends are not avalable, some unconventonal fault locaton technques based on system-wde sparse measurements may have to be used [9-10]. In ths case, fault locaton s estmated by usng measurements recorded from IEDs nstalled n the substatons close to the faulted lne (but not from the ends of the lne) and also usng SCADA measurements from all the substatons near the fault. Performance of the system-wde sparse measurement based fault locaton algorthms depend on fault resstance. The analyss of the mpact of the fault resstance on the senstvty of fault locaton output s crucal for estmatng an accuracy of the output. In [11-1] senstvty of one-end fault locaton methods s analyzed to determne most contrbutng uncertanty factors and nteracton of uncertanty factors. Ths paper explans how accuracy of system wde sparse measurement based fault locaton can be mpacted by fault resstance. An ntutve scheme to choose proper fault resstance range s also proposed. The next secton s focused on descrbng sparse measurement based fault locaton method and the correcton scheme proposed to reduce senstvty to fault resstance. Software mplementaton and case study are dscussed n the subsequent sectons followed by the concluson.

2 II. SPARSE MEASUREMENT METHOD WITH AULT RESISTANCE SENSITIVITY ANALYSIS A. Sparse measurement based fault locaton method The basc dea of transmsson lne fault locaton s to estmate the dstance of the fault pont from any one end of the lne. gure 1. aulted crcut model In g.1 a fault wth resstance R has occurred on pont between two ends (S and R) of a lne secton S-R. Consderng a homogeneous lne, the dstance can be expressed as a functon of the mpedance measured from one end xz L. The above crcut can be solved accurately f voltage and current measurements from both ends are avalable. Installng recordng devces (DRs n our case) on the ends of all the transmsson lnes s not economcal. Although protectve relays exst on every transmsson lne, most of them may stll be electromechancal and they do not have capablty to record measurements. As a result, n some cases t may happen that there are no recordngs at all avalable at lne ends close to a fault. System-wde sparse measurement based fault locaton method can be appled n such nstances [9-10]. In sparse measurement based fault locaton method, phasor measurements from dfferent substatons located n the regon where the fault has occurred are used. The measurements are sparse,.e. they may come from only some of so many transmsson lne ends (substatons) n the regon. Ths method requres synchronzaton of the measurements, whch may be obtaned by usng DRs connected to Global Postonng System (GPS) recevers [13]. Besdes the sparse measurements, the technque also uses short crcut program, whch s ntalzed and tuned wth SCADA PI Hstoran [14], power system model data and measurements assocated wth the tme of the fault occurrence. The method uses waveform matchng technque between the current and voltage phasors calculated from the waveforms recorded n a substaton (nearby the faulted lne) and phasors smulated usng short crcut smulaton of possble fault locatons. A commercal short crcut program tool PSS/E TM 7 s used for short crcut calculaton [15]. The calculated and smulated phasors are compared whle the locaton of the fault s changed n the short crcut program. Ths process s repeated automatcally untl the dfference between measured and smulated values reaches global optmum (mnmum), whch ndcates that the fault locaton used n the short crcut program s the actual one n the feld. The crtera for the mnmal dfference are based on a global optmzaton technque that uses Genetc Algorthm. In ths approach feld-recorded waveforms are used to calculate phasors and they are n turn matched wth the phasors obtaned usng short crcut study. The matchng degree between the recorded and the smulated waveforms can be formulated as [9]: N V I (1) f x, R r V V r I I c V s r I s r 1 1 Where, f x, R : The cost functon usng phasors for matchng c x, R : The fault locaton and fault resstance rv, r I : Weghts for the errors of the voltages and currents respectvely Vs, V r : Smulated and calculated from measurements durng-fault voltages respectvely Is, I r : Smulated and calculated from measurements durng-fault currents respectvely NV, N I : Total number of voltage and current phasors to be matched respectvely : The ndex of voltage or current phasors Ideally when the smulated phasors and phasor calculated from the recorded waveforms match completely, the cost functon should become zero. In practcal soluton, the cost functon s not zero and should be mnmzed usng some mathematcal optmzaton method. To obtan good phasor matchng the fault search range should be extensve. All possble faulty branches and fault resstance should be ncluded n the search range whch maes the search twodmensonal and exhaustve. or a large system, multple searches should be run n parallel whch can be acheved usng populaton based optmzaton methods such as Genetc Algorthm (GA) [16]. The flowchart of ths method s shown n g.. gure. lowchart of sparse measurement algorthm N

3 B. ault resstance compensaton correcton scheme Snce fault resstance s the most uncertan parameter, a correcton scheme to compensate the effect of fault resstance can be proposed. The correcton scheme (shown n g. 3) compensate the effect of fault resstance by usng an optmzaton scheme over the sparse measurement method whch selects the fault locaton and fault resstance par that cause the output to be least senstve to fault resstance varaton. g R where fault locaton s expressed as a functon of fault resstance: x g R () s assumed to be zero mean error Therefore; expected value of x gven g R E x R R s: (3) We can approxmate the true functon g by g usng tradtonal non-parametrc regresson method [17]. Step 3: Calculate senstvty to R : The varance based global senstvty analyss method (ANOVA decomposton) s used [18]. The method s summarzed below: If we consder a determnstc model Z f y Where y s a vector of nput varables y y y y Z s the model output We can decompose nteractons,,..., 1 Z f y nto man effects and 0 j, j (4) 1,,..., 1 1 j... f y, y,..., y f y f f y f y y If each term s chosen wth zero mean 1 0 f y dy 0, y, 1,,..., gure 3. lowchart of corrected sparse measurement algorthm The proposed corrected sparse measurement algorthm can be explaned n the followng step by step method: Step 1: Intalzaton: Generate a populaton of fault resstance values R (1 to N) Step : Sparse measurement based fault locaton: or each R, perform orgnal sparse measurement algorthm and get fault locaton x (1 to N). Therefore a par of x, R s obtaned for N number of cases. Step 3: Estmaton of unnown functon g by nonparametrc method: We have N pars of observatons x R for 1: N, whch are used to estmate the unnown regresson functon Therefore So we can wrte f y, y dy dy 0, y, y, j j j j j f y, y,..., y dy dy... dy 0 1,,..., 1 1 f y dy f0 (5) 0 j, j (6)... f y, y,..., y 1,,..., 1 Z E Z f f y f y y 1 j

4 As terms orthogonal, we can square and ntegrate (6) over and decompose the varance of f ( y) nto terms of ncreasng dmensonalty Z EZ y EZ y, y j... 1 j j E Z y, y, y j Now we can defne senstvty ndces as: E Z y S Z effect of y (7) : rst order ndex. Corresponds to man EZ y, y j Sj : Second order ndex. It measures the Z effect of pure nteracton between any par of factors on the output And so on. So the (7) becomes 1 S S... S (8) j 1,,..., 1 j Therefore n our case, Senstvty of x wth respect to (frst order ndex) s gven by: S R R Ex R (9) x Here we don t now the true g but we can obtan ts estmate g by the tradtonal non-parametrc regresson method. E x R Hence we can estmate by 1 SE x R g g (1) N 1 Where g g R Now we can segment the range of R n some sub-ranges and also determne the senstvty n those ndvdual subranges. Step 4: Determne optmal par of ( x, R ) : rom the dfferent senstvty ndces n dfferent subranges of R, we can choose the sub-range of R that corresponds to least senstvty. Now the optmal par of ( x, R ) should be the par that corresponds to mnmum of msmatch computed n (1). III. IMPLEMENTATION The archtecture of the fault locaton scheme s shown n g. 4. Several commercal pacages are used to mplement ths soluton. The statc power system s modeled usng PSS/E TM 7. To tune the power grd wth pre-fault data, SCADA PI- Hstoran data s used. Now Ex R and x are unnown parameters whch can be estmated from the par of observatons generated. Now x s estmated by ts unbased estmator 1 x S x x (10) N 1 whch s also nown as sample varance of x E x R g R And s equal to whch can be estmated by ts unbased estmator Where g g R 1 E x R N 1 S g g (11) gure 4. Soluton archtecture The detaled data requrements for the mplementaton are:

5 Statc system model data: These nclude power flow system specfcaton data for the establshment of a statc system model (n *.raw format). Power system model data can also be used n saved case format (*.sav) whch s used to extract *.raw data. Event data: These nclude event data captured by recordng devces (DRs here) after occurrence of a fault. The raw DR data s converted to COMTRADE format [19] usng DR Assstant software [0] whch can generate an analyss report (contanng the type of fault and a possble faulted lne) n addton to generatng the COMTRADE fles. The COMTRADE fles contan: o Confguraton fles(*.cfg): nformaton for nterpretng the allocaton of measured data to the equpment (nput channels) for a specfc substaton o Data fles (*.dat): analog and dgtal sample values for all nput channels (descrbed n confguraton fle) n substaton SCADA PI Hstoran data: Ths data reflects real tme changes n power system ncludng the latest load, branch and generator data to tune the statc system model wth the actual pre and post fault condtons. The nomenclature of power system components n all three types of data s dfferent. Correlaton between all three types of data s requred. Substaton nterpretaton fles are prepared to correlate the nomenclature used n DR fles and the one used n PSS/E fle and PI Hstoran data. The nterpretaton fles should be modfed as frequently as needed to reflect the DR confguraton or system model changes. Implementaton of fault locaton software s a four step procedure: A. System ntalzaton Ths s a onetme procedure used to set up the system. Power system statc model data (n *.raw format) s used to extract all the components and construct topology whch wll be used later. B. Pre-process event data The event data captured by DRs should be pre-processed to obtan requred nformaton to be ntegrated wth power system model data. The pre-fault phasor can be calculated usng frst cycle of the recorded waveform. The durng-fault phasor can be calculated usng any fault cycle followng the fault ncepton and pror to fault clearance. The fault ncepton moment s determned from waveforms recorded by DR. or a typcal fault case, several DRs may be trggered and the phasors calculated from the recorded waveforms may lac tme synchronsm whch wll ntroduce phase angle dfference among phasors. Thus tme synchronzaton of the phasors obtaned from dfferent DRs s necessary. The phasors calculated from each DR recordng are synchronzed by rotatng them n reference to the phasors obtaned by the load flow study assumng the angle dfference between the pre- and durng-fault phasor, for the correspondng recorded current or voltage, s fxed. Ths way, all recorded pre- and post-fault phasors are synchronzed usng the same reference. C. Tunng wth real tme data The statc system model may not reflect the operatng condtons of the system when an event s recorded. A tunng wth real-tme power systems s requred. The tunng procedure s done n two steps: Tunng topology: The topology update s performed usng nformaton of the pre-fault breaer status and the pre-fault current magntudes of the montored branches derved from the DR data. It s assumed that a zero magntude (or smaller than 0.01 p.u.) of the current through a montored branch ndcates an outof servce status of the branch. Tunng generaton and load data: The SCADA PI Hstoran data s load, branch and generator data scan (typcally 10 sec nterval) n a perod before and after fault for each substaton where DRs trggered. These data were used to update the system load and generaton. The updated model s saved n a new saved case data (*.sav) whch s used for further smulaton. D. Estmatng fault locaton and evaluatng senstvty to R The fault locaton soluton usng GA s performed n the followng steps. The outer loop optmzaton requres teratons wth dfferent R n a pre-determned range. The ntal populaton for the nner loop optmzaton s chosen randomly for ths one dmensonal (.e. wth one varable x ) optmzaton problem. ault locaton varable can be chosen from a range of zero to the length of the possble faulted lne. Short crcut studes are carred out usng PSS/E and the msmatch value from (1) s evaluated for each of the possble fault locatons. Now by usng three GA operators (selecton, crossover and mutaton) fault posng for next teraton s obtaned. By teratvely posng faults, runnng short crcut smulatons, evaluatng the ftness value, and updatng the fault locaton and resstance, the GA based search engne gudes the search process for a globally optmal soluton for a gven value of R. Now the varance based senstvty analyss method s used to determne senstvty of fault locaton wth respect to the fault resstance n the parttoned sub-ranges of R. The sub-range correspondng to least senstvty s chosen and the mnmum msmatch n that range corresponds to the optmal par of ( x, R ) for that fault. IV. CASE STUDY An actual utlty case study s presented here. In the faulted networ shown n g. 5, a DR nstalled on bus 1 s trggered upon the occurrence of the fault and DR report ndcates the fault s on the lne secton 1-5 whle actually the fault was reported to be 3 mles from bus 8. Wth our algorthm, the fault s recognzed as beng ether n lne secton 6-5 or n lne secton 7-9.

6 Our software yelds much more accurate fault locaton estmaton than what s feasble usng other technques. ault resstance range s chosen as 0 per unt to 0.8 per unt and. ault resstance s changed wthn ts range by ncreasng ts value by per unt n each teraton. gure 5. aulty networ ault resstance vs. fault locaton s plotted n g.6. Senstvty analyss wth respect to fault resstance s performed for several teraton runs and the senstvty ndces correspondng to dfferent sub-ranges of R are presented n Table 1. gure 6. R vs. x plot Ths reflects that for fault resstance n the range of per unt, the sparse measurement based fault locaton method yelds accurate result but after that the algorthm becomes senstve to the choce of fault resstance. Senstvty to R ( S R ) TABLE I. SENSITIVITY TO R OR DIERENT R RANGES Range of R (per unt) Run Run V. CONCLUSION An effcent scheme for reducng the mpact of fault resstance on the sparse measurement fault locaton algorthm for transmsson lne s proposed. The followng are the scheme propertes: It can estmate fault locaton and fault resstance value even f the measurement from the ends of the faulty lne are unavalable. A correcton scheme to reduce the mpact of fault resstance whch s an unpredctable parameter n fault locaton estmaton procedure s proposed. To acheve better accuracy, ths method taes advantage of waveform data recorded by IEDs and archved data measured by SCADA RTUs. The fault resstance compensaton process s automated to allow practcal use n actual power networ applcaton. REERENCES [1] M. Kezunovc, et al, IEEE Gude for Determnng ault Locaton on the Transmsson and Dstrbuton Lnes, IEEE C [] T. Taag, Y. Yamaosh, Y. Yamaura, R. Kondow and T. Matsushma, A new algorthm of an accurate fault locaton for EHV/UHV transmsson lnes: part I ourer transform method, IEEE Trans. Power Appar. Syst. vo. 100, no. 3, pp , [3] T. Taag, Y. Yamaosh, M. Yamaura, R. Kondow, and T. Matsushma, Development of a new type fault locator usng the onetermnal voltage and current data, IEEE Trans. on Power App. and System, vol. 101, no. 8, pp , Aug [4] L. Ersson, M. Saha and G. D. Rocefeller, An accurate fault locator wth compensaton for apparent reactance n the fault resstance resultng from remote end nfeed, IEEE Trans. Power Apparatus and Systems, vol. PAS-104, no., pp , eb [5] V.Terzja, H.-J.Kogln. On The Modelng of Long Arc n Stll Ar And Arc Resstance Calculaton. IEEE Transactons on Power Delvery. 004 July; 19(3): [6] M.Djurc, Z.Radojevc, V.Terzja. Tme Doman Soluton Of ault Dstance Estmaton And Arcng aults Detecton On Overhead Lnes. IEEE Transactons on Power Delvery January; 14(1): [7] A. Johns, S. Jamal, Accurate fault locaton technque for power transmsson lnes, IEE Proceedngs, vol. 137 Pt. C, no. 6, pp , [8] G. Preston, Z. M. Radojevć, C. H. Km, V. Terzja, New settngs-free fault locaton algorthm based on synchronsed samplng, IET Generaton, Transmsson & Dstrbuton, vol. 5, ssue 3, pp ,March 011. [9] M. Kezunovc, Y. Lao, ault locaton estmaton based on matchng the smulated and recorded waveforms usng genetc algorthms, Development n Power System Protecton, Amsterdam, The Netherlands, Aprl 001. [10] M. Kezunovc, P. Dutta, ault locaton usng sparse wde area measurement, CIGRE Study Commttee B5 Annual Meetng and Colloquum, Jeju, October 009. [11] R. Zvanovc, Evaluaton of transmsson lne fault-locatng technques usng varance based senstvty measures, 16 th Power System Computaton Conference, 008. [1] R. Zvanovc and H. B. Oo, A Systematc Approach for Testng Transmsson Lne ault-locatng Technques, Electrcty 007 Conference, The Electrc Energy Socety of Australa (EESA), August 007, Melbourne, Australa. [13] A. Phade, et. al., Synchronzed Samplng and Phasor Measurements for Relayng and Control, IEEE Trans. Power Delvery, Vol. 9, No. 1, pp , January 1994.

7 [14] PI Hstoran, OSISoft, [Onlne] Avalable: [15] PSS E Load low Semens Power Transmsson & Dstrbuton, Inc., PTI, Schenectady, NY, USA. [16] D. E. Goldberg, 1989, Genetc Algorthms n Search, Optmzaton and Machne Learnng, Addson Wesley, Readng, MA. [17] R. L. Euban, 1988, Nonparametrc Regresson and Splne Smoothng, Marcel Dear, N.Y., Basel. [18] I.M. Sobol, Senstvty Analyss for Non-lnear Mathematcal Models, Mathematcal Modellng & Computatonal Experment, Vol. 1, No. 4, pp , [19] IEEE Inc., IEEE standard Common ormat for Transent Data Exchange (COMTRADE) for power systems, IEEE Std C , [0] Test Laboratores Internatonal, Inc.: DR Assstant - Software for Automated Analyss and Archval of DR records wth Integrated ault Locaton Calculaton [Onlne]. Avalable: