A NEURO-FUZZY APPROACH FOR THE FAULT LOCATION ESTIMATION OF UNSYNCHRONIZED TWO-TERMINAL TRANSMISSION LINES

Transcription

1 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 A NEURO-FUZZY APPROACH FOR THE FAULT LOCATION ESTIMATION OF UNSYNCHRONIZED TWO-TERMINAL TRANSMISSION LINES Ramadon Syahputra Department of Electrcal Engneerng, Faculty of Engneerng, Unverstas Muhammadyah Yogyakarta, Yogyakarta, 5583, INDONESIA ramadon@umy.ac.d ABSTRACT Overhead transmsson lne s an element of electrcal power systems that are most frequently experenced short crcut faults compared to other power system elements. Short crcut faults on overhead transmsson lne cause a relatvely large current and therefore can damage mechancally the electrcal equpment connected to the system. The protecton system s essentally needed n ths stuaton. In addton, t takes a pece of equpment that can detect the locaton of fault n order to expedte the repar process, especally f the fault s permanent state. In ths paper, a neuro-fuzzy aprroach for short crcut fault locaton estmaton whch uses data from both ends of overhead transmsson lne s descrbed. The approach utlzes the advantages of dgtal relayng whch are avalable today. The unsynchronzed data of fault voltages and currents at two-end of overhead transmsson lne s appled n ths technque. The accurate fault locaton estmaton technque has rrespectve of source mpedances, fault resstances, fault types, and load currents. Smulaton of short crcut fault of transmsson lne has done by usng EDSA software. Short crcut currents and voltages from both ends of overhead transmsson lne have used to nput data of neuro-fuzzy method n Matlab program. Smulaton results demonstrate the accuracy of the method. The results shows that the lowest estmaton error for sngle phase to ground fault wth the varaton of fault resstances of 0 ohms, 0 ohms, 50 ohms, and 00 ohms, respectvely, s %, whle the hghest estmaton error s %. KEYWORDS Neuro-Fuzzy, Two-termnal fault locaton algorthm, transmsson lne, unsynchronzed samplng.. INTRODUCTION The transmsson lne s an element of electrcal power systems that are most frequently experenced short crcut faults compared to other electrcal power system elements, wth the percentage of probablty of occurrence of 50% []. There are four types of short crcut faults on a transmsson lne,.e., sngle phase to ground fault, phase to phase fault, double phase to gound fault and symmetrcal three-phase fault. Short crcut faults on overhead transmsson lne cause a relatvely large current and therefore can damage mechancally the electrcal equpment connected to the system. In addton, t takes a pece of equpment that can detect the locaton of fault n order to expedte the repar process, especally f the fault s permanent. The contnuous and relable electrcal energy supply s the am of electrcal power system operaton. The short crcut faults on the lne must be estmated accurately to allow mantenance operators to arrve at the scene and repar the faulted transmsson lne part as soon as possble. Geographcal layout and rugged terran make some sectons of electrcal power transmsson lnes dffcult to reach; DOI : 0.52/jcst

2 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 therefore, the robustness of fault locaton estmaton under a varety of power system operatng constrants and fault condtons s an mportant requrement. Therefore, t s mportant to know where the locaton of a faulted transmsson lne secton has been occured. Development of the applcaton of dgtal and mcroprocessor-based power protecton systems has been motvated for fault locaton technque researchers n the last few decades [2]-[4]. However, accuracy of the dstance to the fault determnaton from a power substaton s affected by several stochastc factors. There are some factors whch affectng the accuracy of fault estmaton,.e., fault resstance, equvalent mpedances of transmsson lne, effect of the load varaton and mprecsons of the transmsson lne parameters measurements [4]-[6]. Therefore, the accuracy of fault dstance determnaton from both ends of transmsson lne may be nsuffcent. Accurate locaton of faults on overhead electrcal power transmsson lnes for the nspectonrepar purpose s of vtal mportance for utlty staff and operators for expedtng servce restoraton, and thus to reduce outage tme, operatng costs and customer complans. Durng the last decade a number of fault locaton algorthms have been developed, ncludng the transentstate approach, steady-state phasor approach, the travelng-wave approach, the lumped parameter approach, and the dfferental equaton approach [7], as well as two-end [8] and one-end [9] algorthms. In the last category, synchronzed [0] and non-synchronzed [] samplng technques are used. However, two-termnal data are not wdely avalable on the protecton equpment. Fault locaton determnaton methods usng the measured voltage and current at one termnal channel s stll nadequate because t does not take nto account the current channel nterference from the other termnal. In [2] usage of synchronzed measurements of currents and voltages from all two termnals of transmsson lne has been consdered. They have used the dstrbuted parameter models of the overhead transmsson lne sectons. The approach assures hgh accuracy of fault locaton estmaton, and the faulted transmsson lne secton s also rely on ndcated [2]. Then the use of three-termnal unsynchronzed measurements of short crcut fault currents and voltages has been consdered n [3]. They have used the lumped parameter models of the overhead transmsson lne sectons. It was mpled that the error estmaton resultng from such smplfcaton s mnmzed due to the redundancy of the fault locaton equatons. Yet another utlzaton of three end unsynchronzed measurements has been proposed n [4], where exchangng the mnmal amount of nformaton between the lne termnals over a protecton channel was consdered. After that, the development of fault locaton technques for three termnal transmsson lnes that utlzng only from two-end synchronzed measurements of currents and voltages has been studed n [5]-[6]. From a practcal vewpont, t s desrable for equpment to use only one-termnal data. The one-end algorthms, n turn, utlze dfferent assumptons to replace the remote end measurements. Most of fault locaton algorthms are only based on local measurements. The use of fuzzy logc based n electrcal power system study has been became an nterestng research n two last decades [7]. Currently, the most wdely used method of overhead transmsson lne fault locaton s to determne the apparent reactance of the lne durng the tme that the short crcut fault current s flowng and to convert the ohmc result nto a dstance based on the parameters of the lne. It s wdely recognzed that ths method s subject to errors when the fault resstance of the transmsson lne s hgh and the lne s fed from both ends (substaton), and when parallel crcuts exst over only parts of the length of the faulty lne. In ths paper, a neuro-fuzzy approach for short crcut fault locaton estmaton, whch utlzes unsynchronzed measurements of currents and voltages at two-termnal of transmsson lne s proposed. The measurements has used to nput data for tranng process n neuro-fuzzy system. The lumped parameter n medum transmsson lne model s strctly used. The approach can be appled n dgtal dstance protecton whch s avalable today n electrcal power system. The method allows for accurate estmaton of short crcut fault locaton rrespectve of fault types, 24

3 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 fault resstance, load currents, and source mpedances. The powerful of both EDSA software and Matlab software have used n ths study. 2. FUNDAMENTAL THEORY 2.. Faults n Transmsson Lnes Transmsson lnes s one of the man component n electrc power systems. Transmsson lne s a component of electrcal power systems that are most frequently experenced faults compared to other power system elements. Faults n power systems are bascally classfed as shunt faults and seres faults [8]. Both types can be balanced fault or unbalanced fault. Faults on a transmsson lne can occur as whether sngle or smultaneous faults. Most of faults n transmsson lnes are sngle lne to ground type. Therefore, the sngle lne to ground fault wth varous fault resstances s used n ths study. Smultaneous faults on a transmsson lne consst of a combnaton of the same or dfferent types of faults. The preferred model for fault calculaton s the nodal approach n the frequency doman wth transmsson lne symmetrcal components or n the tme doman wth space phasors or ts components. Conventonal technques for fault locaton calculaton are usually based on admttance equatons n admttance form. Another general problem s the treatment of smultaneous lne faults. Each fault s characterzed by the so-called boundary condtons regardng the currents and voltages at the fault locaton of transmsson lne. In an electrcal power system comprsng of any varous nteractng elements, there always exsts a possblty and probablty of faults. The emerge of large power generatng statons and hghly nterconnected power systems va any overhead transmsson lnes makes early fault detecton and rapd equpment solaton mperatve to mantan the stablty of the system. Faults on overhead transmsson lnes need to be detected rapdly, located accurately and repared as soon as possble. Fault detector module of a transmsson lne protectve devce can be used to start other relayng modules. The detectors provde an addtonal way of securty n a power protecton relayng applcaton as well, and the locaton of short crcut fault must be determned. Besdes beng used to accurately locate a fault, such an method can be used for automated fault analyss. Any occurrence of a short crcut fault should be detected and cleared by the power protectve relayng devces. A power protectve relayng operaton analyss s requred f an assessment of ts performance s needed. In order to perform the analyss, one has to have a reference method wth whch to compare the dgtal relay operaton. The fault locaton technque that can provde both fault type classfcaton and accurate locaton s an deal reference for the robust dgtal protectve relayng operaton. The technque can be ncorporated nto a fault analyss automatcally by provdng hgh speed nformaton of the fault type and fault locaton. Ths s mportant nformaton for determnng f a power protectve relay has operated correctly snce the relay s also supposed to determne both type and locaton of fault. The locaton of fault determned by the dstance relay does not have to be too accurate snce t only has to determne the zone of the occurrence of fault. The approprate locaton of the fault provded by the fault locaton technque s more accurate and s needed by power operators Neuro-Fuzzy Method Durng the last two decades adaptve neuro-fuzzy approach has been became a popular method n control area. In ths part, a bref descrpton of the adaptve neuro-fuzzy nference system (ANFIS) prncples s gven whch are refered to [9]. The fundamental structure of the type of 25

4 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 fuzzy nference system (FIS) could be seen as a model that maps nput characterstcs to nput membershp functons. After that, t maps all membershp functon as nput to rules and rules to a set of characterstcs of FIS output. On the last step, FIS maps characterstcs of output to membershp functons as output, and the membershp functon as output to a decson assocated wth the output. As can be seen that FIS has been stated only non-arbtrary membershp functons that were chosen arbtrarly. Fuzzy nference system (FIS) s only used to modelng systems whose the structure of fuzzy rule s essentally predetermned by the operator nterpretaton of the varable characterstcs n the model. However, t cannot be dstngush what the FIS membershp functons should look lke smply from the data for some stuatons. Parameters of FIS could be chosen so as to talor the membershp functons to the nput and output data rather than choosng the parameters assocated wth a gven membershp functon arbtrarly n order to account for these types of varatons n the values of data. Therefore, the necessty of an adaptve propertes n fuzzy nference system becomes obvous. The adaptve neuro learnng concept works smlarly to the artfcal neural networks. Neuroadaptve learnng technques provde a method for the fuzzy modelng procedure to learn nformaton about a data set. It computes the membershp functon parameters that best allow the assocated fuzzy nference system to track the gven nput and output data. A network-type structure smlar to that of an artfcal neural network can be used to nterpret the nput and output map so t maps nputs through nput membershp functons and assocated parameters, and then through output membershp functons and assocated parameters to outputs. Through the learnng procedure, parameters whch assocated wth the membershp functons wll changes. The parameters computaton s facltated by a vector of gradent. The vector of gradent determne a crteron of how well the fuzzy nference system (FIS) s modelng the nput and output data for a gven parameters set. Whle the vector of gradent s obtaned, several optmzaton procedures can be used n order to control the parameters to reduce some error measure (ndex of performance). The error measure s comonly defned by the sum of the squared dfference between desred and actual outputs. ANFIS uses a combnaton of back propagaton procedure and least squares estmaton for membershp functon parameter estmaton. 26

5 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 µ µ A B w f = p x + q y + r x y µ µ A 2 B 2 w 2 f 2 = p 2 x + q 2 y + r 2 x y wf + w f f = w + w Fgure. Sugeno s fuzzy logc model Layer Layer 2 Layer 3 Layer 4 Layer 5 x y x A A 2 w N w w f f y B B 2 w 2 N w 2 x y w 2 f 2 The suggested ANFIS has several propertes: Fgure 2. The archtecture of the ANFIS.. The ANFIS output s Sugeno-type of zero-th order. 2. ANFIS has only a sngle output whch obtaned usng defuzzfcaton process of weghted average. All output membershp functons are constant. 3. It has no rule sharng. The number of rules must be equal to the number of output membershp functons. 4. It has unty weght for each rule. 27

6 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 Fgure shows Sugeno s fuzzy logc model. The ANFIS archtecture s shown n Fgure 2. The archtecture comprsng by nput, fuzzfcaton layers, nference unt and defuzzfcaton layers. The ANFIS archtecture can be desrpted as consstng of N neurons n the nput layer and F membershp functons for each nput, and F*N neurons n the fuzzfcaton layer. The nference unt and defuzzfcaton have FN rules wth FN neurons, whle the output layer has one neuron. For smplcty, t s assumed that the fuzzy nference system under consderaton has two nputs x and y and one output z as shown n Fgure 2. For a zero-order Sugeno fuzzy model, a common rule set wth two fuzzy f-then rules s the followng: Rule : If x s A and y s B, Then f = r () Rule 2: If x s A2 and y s B2, Then f2 = r2 (2) Here the output of the -th node n layer n s denoted as O n, : Layer. Every node n ths layer s a square node wth a node functon: O = µa (x), for =, 2, (3) or, O = µb -2 (y), for = 3, 4 (4) where x s the node- nput, and A s the label of lngustc terms (bg, low, etc.) assocated wth ths node functon. O s the A membershp functon. O specfes the degree to whch the gven x satsfes the A. Usually µa(x) s chosen to be bell-shaped wth maxmum equal to and mnmum equal to 0, such as the generalzed bell functon: μ A (x) = + x c a 2b (5) Parameters n ths layer are referred to as premse parameters. Layer 2. Each node n layer 2 s labeled by Π whch multples the ncomng data and sends the product out. For nstance, 2 O = w = µa(x) x µb(y), =, 2. (6) Each node output represents the frng strength of a rule. Other T-norm operators whch shows generalzed AND can be used n layer 2. Layer 3. Every node n ths layer s a crcle node labeled N. The -th node calculates the rato of the -th rule s frng strength to the sum of all rules frng strengths: O w 3 = w =, =, 2. (7) w + w2 The outputs of layer 3 wll be mentoned as normalzed frng strengths. Layer 4. Every node n ths layer s a square node wth a node functon: 28

7 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February O = w f = w (p x + q y + r ) (8) w s the layer 3 output, whle {p, q, r } s the set of parameter. All parameter n ths layer wll be mentoned as consequent parameters. Layer 5. The sngle node n ths layer s a crcle node labeled Σ that computes the overall output as the summaton of all ncomng sgnals,.e., O 5 = w f (9) 2.3. Unsynchronzed Samplng The procedure of short crcut fault locaton estmaton n ths study s use two termnals of electrcal power transmsson lne as shown n Fgure 3 []. V A I A I B V B Z A mz (-m)z Z B V A E BUS A I f V f BUS B V B E Fgure 3. Short crcut fault n transmsson lnes. Transmsson lne as shown n Fgure 3, both voltage and current phasors from protected two termnals of the lne are requred n ths procedure, but unsynchronzed. As can be seen n Fgure 3, the method wll estmate fault dstance m from two ends of overhead electrcal power transmsson lne. The fault voltages and currents from bus A and bus B of transmsson lne are not synchronzed, whle δ s angle synchronzaton. For example, the voltage at bus A and bus B can be wrtten as follow: V A = V A m + ; V B = V B m (0) where α m and β m are measured angle from two ends respectvely, and δ s the synchronzaton phasor angle between bus A and bus B. The smlar equaton can be wrtten for current phasors. Hence, equaton (0) has become: V A e j V B + Z I B = mz (I A e j + I B ) () The unknown components of equaton ( ) are dstance fault m and complex number δ = e jδ. Equaton () can be separated nto real part and magnary part to forms the new two equatons as follow: Re(V A )sn + Im(V A )cos - Im(V B ) + K 4 = m(k sn + K 2 cos + K 4 ) (2) Re(V A )cos - Im(V A )sn - Re(V B ) + K 3 = m(k cos - K 2 sn + K 3 ) (3) 29

8 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 Coeffcents of K, K 2, K 3, and K 4 n equaton (2) and equaton (3) can be defned as follow: K = R Re(I A ) X Im(I A ) (4) K 2 = R Im(I A ) + X Re(I A ) (5) K 3 = R Re(I B ) X Im(I B ) (6) K 4 = R Im(I B ) + X Re(I B ) (7) Then, the equaton wth unknown angle δ s formed. As rearranged the equatons above, then the new equatons are resulted as follow: a sn + b cos + c = 0 (8) where, a= K 3 Re(V A ) K 4 Im(V A )-K Re(V B ) K 2 Im(V B )+K K 3 +K 2 K 4 (9) b= K 4 Re(V A ) K 3 Im(V A ) K 2 Re(V B ) +K Im(V B )+K 2 K 3 K K 4 (20) c= K 2 Re(V A )-K Im(V A )-K 4 Re(V B )+K 3 Im(V B ) (2) From the equaton ( 8) can be seen that angle δ (synchronzaton angle) s unknown. The unknown one can be found by usng Newton-Raphson teratve algorthm. The equaton for teratvely to count the angle δ (n radan) s: k+ = k F( k ) F'( ) k (22) Iteratve process wll be stopped when the dfference between two end values that smaller than 4 the stated float s acheved, for example: 0 k + k <. The procedure has the quadrate convergence and needs ntal value for runnng the process. If voltage angle of two termnals of transmsson lne are zero respectvely, then angle δ be apparent angle between two voltages and ndependent to synchronzaton error. As the synchronzaton has known, fault dstance from one of two ends of lne can be calculated from equaton (2) and equaton (3). From equaton (2): Re( VA)sn + Im( VA)cos Im( VB ) + K m = K sn + K cos + K (23) If we use the equaton ( 3), then fault dstance m from a transmsson lne termnal can be calculated as follow: Re( VA)cos Im( VA)sn Re( VB ) + K m = K cos K sn + K (24) Equaton (23) and (24) can be used to all types of short crcut faults. Ths method can apply on overhead power transmsson lne wth mult-phase cases. The amount of approprate phase can be used n ths method. 30

9 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 And, error estmaton can be calculated by the equaton below [9]: actual locaton estmated locaton Error estmaton(%) = 00% lne length (25) 3. SIMULATION RESULTS The procedure of ths research s shown n Fgure 4. Smulaton of short crcut fault of transmsson lne has done by usng EDSA software. Short crcut currents and voltages from both ends of overhead transmsson lne have used to nput data of neuro-fuzzy method n Matlab envronment. It s used as the man engneerng tool for performng modellng and smulaton of electrc power systems, as well as for nterfacng the user and approprate smulaton programs. MATLAB has been chosen due to avalablty of the powerful set of programmng tools, sgnal processng, numercal functons, and convenent user-frendly nterface. In ths specally developed smulaton envronment, the evaluaton procedures can be easly performed. We have used Fuzzy logc Toolbox of MATLAB to develop the ANFIS model wth 2 nputs and sngle output as gven n Fgure 2. The smulatons were performed by an Intel Pentum core(tm) 2 duo CPU,.80 GHz, 4 GB RAM. For evaluatng the performance of the proposed algorthm, the author adopts EDSA software for fault data generaton and Matlab for neuro-fuzzy algorthm mplementaton. Power system shown n Fgure 3 has selected for the studes reported n ths paper. Bus A and bus B were consdered to be connected by 50 kms, 500 kv transmsson lnes [2]. Two equvalent power systems consdered to be connected to bus A and bus B. Electrcal parameters of transmsson lnes and equvalent power sources are gven n Table I. TABLE I ELECTRICAL PARAMETERS OF TRANSMISSION LINE IN THE STUDY Impedances Components Postve sequence (Z ) Negatve sequence (Z 2 ) Zero sequence (Z 0 ) Transmsson lnes 53,090 86,07 53,090 86,07 73,683 72,96 Power source A 4,00 89,03 3,50 88,90,44 45 Power source B 00,020 88,85 75,027 88,47 25,072 85,65 3

10 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 Start Lterature study Create the transmsson lne model n EDSA Create the neuro-fuzzy fault locator n Matlab Examne the short crcut fault on transmsson lne Use the fault currents and voltages as nput data of neuro-fuzzy program n Matlab Examne neuro-fuzzy program for fault type and fault locaton estmaton Examne the other faults? Yes No Analyze the performance neuro-fuzzy method Concluson Fnsh Fgure 4. Procedure of the research. In order to estmate the fault locaton of transmsson lne accurately, the fuzzy systems are traned wth a separate ANFIS structure and sutable off-lne data. The man steps of the procedure are (see Fg. 4): Step : Create the archtecture of fuzzy nference system (FIS) n Matlab envronment. The archtecture consst of two nputs (.e. voltages and currents fault measured n the locator end of transmsson lne) and one output. Step 2: Determne the membershp functons for ANFIS nput and output, recpectvely, and then defne the If-Then rules. In ths work, gbell membershp functon has been choosen for each nput and output of FIS archtecture. Step 3: Collectng or producng sutable nformaton (data) to tran ANFIS. The data for tranng process should have same form and the varous condtons of a real power systems ncluded. A power system smulaton usng EDSA has been carred for achevng the sutable data. 32

11 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 Step 4: Tranng process. The sutable data whch are collected n step ( 3) are presented to network and adaptve nodes are adjusted. Ths presses wll be stopped when error meet proposed goal. The nodes of adaptve propertes update after entre patterns have been presented to network. Step 5: Testng process. In tranng procedure, the fault locator should be gven an acceptable output for unseen data. If test pattern output reached an acceptable range, fuzzy rule s adjusted n the best condtons. Fgure 5 shows the membershp functon of nput varable Voltage, whle Fgure 6 shows the membershp functon of nput varable Current. The voltage and current nput varables have addressed as the the fault varables of transmsson lne under consderaton. Fgure 7 shows the tranng data of ANFIS and the ANFIS output for 20 epochs. For pre-fault calculatons, all transmsson lnes were modelled by equvalent p networks and all loads were consdered to be constant power loads. For calculatng fault currents and voltages on the ncepton of a fault, the selected lne has modelled by two equvalent p networks, one for the secton from bus A to the fault and the other for the secton from bus B to the fault. For examnng the fault dstance technque, t was assumed that dgtal dstance relays have been provded at the lne termnals on bus A and bus B. It was also assumed that these dgtal dstance relays measure fundamental frequency voltages and currents from sampled data. Fgure 5. Membershp functon of nput varable Voltage. Fgure 6. Membershp functon of nput varable Current. 33

12 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February Tranng Data ANFIS Output Fgure 7. Tranng data of ANFIS and the ANFIS output. TABLE II FAULT DISTANCE ESTIMATION FOR SINGLE PHASE-TO-GROUND FAULT Actual Dstance (km) Estmaton Error (%) R f = 0 Ω R f = 0 Ω R f = 50 Ω R f = 00 Ω 0 0, , , , , , , , , , ,0225 0, , , , , , , , , , ,0430 0, , , , , , , , , , , , , , , , , , , , , , In order to test the powerful of neuro-fuzzy method n ths research, smulaton of a sngle phase to ground fault has done n EDSA envronment. The fault has occured on the selected locaton of transmsson lne. Buses A and B and some locatons,.e. 5 kms, 30 kms, 45 kms, 60 kms, 75 kms, 90 kms, 05 kms, 20 kms, and 35 kms, were chosen as fault locatons. Fault resstances were vared from 0 ohms, 0 ohms, 50 ohms, to 00 ohms. Fundamental frequency voltages at bus A and bus B and lne currents were calculated and provded to the fault locaton program n 34

13 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 Matlab as nputs. Fault locaton of a transmsson lne studes for sngle phase to ground fault are reported n ths paper, as shown n Table II and Fgure 8. Estmaton Error as a Functon of fault Dstance for Sngle Phase to Ground Fault.2 Estmaton error (%) Fault dstance (km) Rf = 0 ohm Rf = 30 ohms Rf = 70 ohms Rf = 0 ohms by Sachdev & Agarw al Rf = 0 ohms Rf = 50 ohms Rf = 00 ohms Fgure 8. Estmaton error as a functon of fault dstance for sngle phase to ground fault on transmsson lne. Table II lsts the estmaton error of fault locatons for a sngle phase to ground fault wth fault resstances of 0 ohms, 0 ohms, 50 ohms, and 00 ohms, respectvely. The estmaton errors expressed as percentages of the lne length are shown n Fgure 5. The results ndcate that dstances of faults estmated by the proposed method are substantally more accurate than the dstances estmated by Sachdev and Agarwal [22]. When a sngle phase to ground fault occurs n bus A (0 kms dstance) wth fault resstance of 0 ohms, the estmaton error s %. Ths error value s the smallest estmaton error n the study. As can be seen n Table II and Fgure 8 that the hghest short crcut fault estmaton error for sngle phase to ground fault s % at dstance of 50 kms wth fault resstance of 00 ohms. 4. CONCLUSIONS Ths paper has proposed a neuro-fuzzy approach that estmates the dstance of a transmsson lne short crcut fault from relay locatons usng unsynchronzed fundamental frequency voltages and currents measured at the two ends of transmsson lne. In ths paper, a neuro-fuzzy aprroach for short crcut fault locaton estmaton whch uses data from both ends of overhead transmsson lne s descrbed. The approach utlzes the advantages of dgtal relayng whch are avalable today. The accurate fault locaton estmaton algorthm has rrespectve of source mpedances, fault resstances, fault types, and load currents. Smulaton of short crcut fault of transmsson 35

14 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 lne has done by usng EDSA software. Short crcut currents and voltages from both ends of overhead transmsson lne have used to nput data of neuro-fuzzy method n Matlab program. Smulaton results demonstrate the accuracy of the method. The results shows that the lowest estmaton error for sngle phase to ground fault wth the varaton of fault resstances of 0 ohms, 0 ohms, 50 ohms, and 00 ohms, respectvely, s %, whle the hghest estmaton error s %. ACKNOWLEDGEMENTS The author would lke to thank profusely and the hghest apprecaton for DIKTI (the Drectorate General of Hgher Educaton) Mnstry of Educaton and Cultural Affars, Republc of Indonesa, for havng funded ths research. REFERENCES [] Ram, B. and Vshwakarma, DN, "Power System Protecton & Swtchgear", pp. 3-6, McGraw-Hll Pub. Co. Ltd., New Delh, 995. [2] A.M. Borbely and J.F. Kreder, Dstrbuted Generaton: The Power Paradgm for the New Mllennum, CRC Press, Washngton D.C., 200. [3] T. Takag, Y. Yamakosh, M. Yamuaura, R. Kondow, and T. Matsushma, Development of a New Type of Fault Locator Usng One Termnal Voltage and Current Data, IEEE Trans. On Power Apparatus and System, vol. PAS-0, No 8, pp , Aug [4] Erksson, L., S.D. Rockfeller and M. Saha, Accurate Fault Loc. wth Compensaton for Apparent Reac. n the Fault Resstance Resultng from Remote End n Feed, IEEE Transactons on Power Aparatur and System, PAS-04, No 2, 985. [5] D. Novosel, D.G. Hart, M.M. Saha, and S. Gress, Optmal fault locaton for transmsson system, ABB Revew 8, pp , 994. [6] A. Sauhats and M. Danlova, Fault Locaton Algorthms for Super Hgh Voltage Power Transmsson Lnes, n Proc. Of IEEE Bologna Power Tech Conf., [7] A. Sauhats and M. Bockarjova, Algorthms, Means and Tools of Fault Locaton on Transmsson Lnes, Proc. EPE-PEMC'2004, Rga, Latva, [8] Lan, B and M.M.A. Salama, An Overvew of Dgtal Fault Locaton Alg. For Transmsson Lne Usng Transent Waveforms, Electrc Power System Research, Vol. 29, No., pp. 7 25, 994. [9] L.B. Sheng and S. Elangovan, A fault locaton algorthm for transmsson lnes, Electrc Machnes & Power Syst., Vol. 26, No. 0, pp , 998. [0] Q. Zhang, Y. Zhang, W. Song, and Y. Yu, Transmsson lne fault locaton for phase-to-earth fault usng onetermnal data, IEE Proc. Trans. Dstrbuton., Vol. 46, No. 2, pp. 2 24, 999. [] Kezunovc, M and J. Mrkc, An Accurate Fault Locaton Alg. Usng Sync. Samplng, Electrc Power System Research, Vol. 29, No. 3, pp. 6 69, 994. [2] Novosel, D., E. Udren, J. Gartty and D.G. Hart, Unsynchronzed Two Termnal Fault Locaton Estmaton, IEEE Transactons on Power Delvery, Vol., No., pp , 996. [3] Aggarwal, R.K., A.T. Johns, D.V. Coury and A. Kalam, A Practcal Approach to Accurate Fault Locaton on EHV Teed Feeders, IEEE Transactons on Power Delvery, Vol. 8, pp , July 993. [4] A.A. Grgs, D.G. Hart, and W.L. Peterson, A new fault locaton technque for two-and threetermnal lnes, IEEE Trans. Power Delvery, Vol. 7, No., pp , January 992. [5] Tzouvaras, D.A., G. Benmmouyal and J. Roberts, New mult-ended fault locaton desgn for two- or three-termnal lnes, Proc. 7th Int. IEE Conf. on Dev. n Power System Protecton, pp , Aprl 200. [6] Ln, Y., C. Yu and C. Lu, A New Fault Locator for Three Termnal Transm. Lnes Usng Two Termnal Sync. Volt. And Cur. Phasors, IEEE Transactons on Power Delvery, Vol. 7, No.3, pp , July [7] R. Syahputra, Fuzzy Multple Objectve App. for the Improvement of Dstrbuton Net. Effcency by Consderng Dstrbuted Generaton, IJCSIT, Vol 4, No 2, Aprl

15 Internatonal Journal of Computer Scence & Informaton Technology (IJCSIT) Vol 5, No, February 203 [8] Brahma, S.M., Fault Locaton Scheme for a Mult Termnal Transm. Lne Usng Synch. Voltage Measurements, IEEE Transactons on Power Delvery, Vol. 20, No. 2, pp , Aprl [9] Jang, J.S.R., 993, "ANFIS: Adaptve-Network-based Fuzzy Inference System", IEEE Trans. Syst., Man, Cybern., 23, , June. [20] B. R. Oswald and A. Panosyan, A New Method for the Computaton of Faults on Transmsson Lnes, IEEE PES Transmsson and Dstrbuton Conference and Exposton Latn Amerca, Venezuela, [2] Izykowsk, J., M.M. Saha, E. Rosolowsk, P. Balcerek and M. Fulczyk, A Fault Locaton Method for the Applcaton wth Current Df. Relays of Three Termnal Lnes, IEEE Transactons on Power Delvery, Vol. 22, No. 4, pp , October [22] M.S. Sachdev and R. Agarwal., A Technque for Estmatng Overhead Transmsson Lne Fault Locatons from Dgtal Impedance Relay Measurements, IEEE Transactons on Power Delvary, Vol. 3, No., pp. 2-29, 988. AUTHOR Ramadon Syahputra receved B.Sc. degree n Electrcal Engneerng from Insttut Teknolog Medan and M.Eng. degree from the Electrcal Engneerng Department, Engneerng Faculty, Gadjah Mada Unversty, Yogyakarta, Indonesa, n 998 and 2002, respectvely. He was wth the Electrcal Engneerng Department, Engneerng Faculty, Unverstas Muhammadyah Yogyakarta (UMY), Indonesa. Hs research nterests nclude computatonal of power system, artfcal ntellgence n power system, power system operaton, power system control, power qualty, dstrbuted generaton, and renewable energy. 37