Estimation of an L-G Fault Distance of an Underground Cable Using WNN

Size: px

Start display at page:

Download "Estimation of an L-G Fault Distance of an Underground Cable Using WNN"

Crystal Mitchell
5 years ago
Views:

1 Iteratioal Joural of Scietific ad esearch Publicatios, Volume, Issue, February ISSN Estimatio of a L-G Fault Distace of a Udergroud Cable Usig WNN Biswariya Chatteree Deartmet of Electrical Egieerig, Future Istitute of Egieerig & Maagemet, West Begal, Idia Abstract- I is aer we have develoed a ree layer Wavelet Neural Networ (WNN for L-G fault locatio of a udergroud cable. Bac Proagatio (BP algorim is used for traiig of e etwor. Simulatio result ad grahs shows e effectiveess of WNN as a fault locator etwor. x x t, s ψ ψ v v y Idex Terms- Wavelet eural etwor, bac roagatio algorim, udergroud cable system, L-G fault locatio E I. INTODUCTION lectrical eergy trasmissio system is demadig safer ad more reliable eergy services. Udergroud cable is oe of e best choices for safer ad reliable ower trasmissio. Alough udergroud cables rovide higher reliability a overhead system, i case of isulatio failure it is very difficult to locate faults. To miimize such difficulties i udergroud system, differet fault idetificatio schemes are beig evolved to locate e fault for roer remedial measures []. Wavelet Neural Networ (WNN is a alterative to feed forward eural etwor for aroximatig arbitrary o-liear fuctios. The WNN maes e best use of e good localizatio roerties of wavelet trasformatio [] ad combie wi e self-learig ability of eural etwors [3]-[4]. Therefore WNN has better aroximatio ad error correctio abilities. I is reset wor, a ree layer WNN has bee develoed to estimate e cable fault (L-G distace. A case study of fault currets was carried out for a udergroud cable for various fault locatio. The fault curret ad fault distace was chose as iut ad outut arameter resectively. The etwor was traied usig bac-roagatio algorim [5]. After e traiig has bee doe, e etwor has bee tested wi e tested data. Exerimetal grahs show at WNN ca redict e fault locatio accurately. II. WAVELET NEUAL NETWOK The structure of a ree-layer Wavelet eural etwor [], [6]- [7] wi multile iuts ad multile oututs is show i Fig x l i,,...l t, s t 3, s 3 ψ 3 ψ m,,...m v m v m Figure : Wavelet Neural etwor,,... As illustrated i Fig e etire etwor cosists of ree layers. First layer is called iut layer which accets e iut data. The umber of iut odes deeds o e umber of iut arameters. The secod layer is called hidde layer where a moer wavelet fuctio is selected as a activatio fuctio which is Morlet wavelet for is wor. Morlet wavelet fuctio is exressed as follows [7]: ψ ( x cos(.75x e ( x ( Outut of e WNN is calculated from e equatio ( m l xi t y ( x [ v * ( ψ ( ] s Where t, s i v ( deotes traslatio ad dilatio arameter resectively. is e weight arameters betwee hidde ad outut euros. The weight arameters betwee iut ad hidde y, x odes are assumed to be uity. i rereset ad i outut ad iut vector resectively. Durig e traiig sessio of e etwor, e iut atter y y causes outut resoses i each layer ad hece at e outut layer is obtaied. The differece betwee estimated ad desired outut yields a error. This error is to be miimized by feedig iformatio forward ad feedig error bacward. The error fuctio is called sum square error [7] which is described i e equatio (3.

2 Iteratioal Joural of Scietific ad esearch Publicatios, Volume, Issue, February ISSN E where, ts d [ d ( x y ( x, y ( x outut ode for e ts (,,..., ts ( x] (3 are e desired ad real outut of e traiig data resectively ad is e total umber of traiig sets. The equatio (3 is e obective fuctio which is miimized usig bac roagatio algorim. O doig so a learig techique is adoted to reduce e learig time. Accordig to is techique v t, s, modificatio of differet etwor arameters viz. are doe usig e followig equatios (4, (5, (6 resectively. v t E ( c v ( c. µ. v ( c] v (4 E ( c t ( c. µ. t ( c] t (5 E s ( c s ( c. µ. s ( c] s (6 η, µ where, are e learig rate ad mometum factor resectively. III. UNDEGOUND CABLE MODEL The schematic diagram of a L-G fault of a udergroud cable i e distributio substatio usig a m log udergroud HT buried cable to distribute ower rough a V / 45 V distributio trasformers is show i Fig [8]. V Source V Cable F Distributio Substatio V / 45 V Y B A B MATS Cable Leg (L m Figure : Schematic diagram of LG fault of a udergroud cable Fault curret i -hase [9] ca be foud out by equatio (7 I 3 I 3 E E, where, I is e re-fault voltage ad ositive sequece, comoet of fault curret resectively ad is e ositive, egative ad zero sequece imedace resectively. (7, IV. KV SYSTEM DATA The V system as show i Fig has ree uits, viz. source, V distributio cable ad eutral ad earig system. The corresodig data [8] of each uit are give i Table I, II, ad III resectively.

3 Iteratioal Joural of Scietific ad esearch Publicatios, Volume, Issue, February 3 ISSN Table I: Source voltage (Volts & imedace (Ω Number Lie voltage V V Vh Sigle hase source voltage 635 V 3 Sigle hase fault level S MVA 4 Positive sequece source imedace (Ω V s.. S 5 Negative sequece source imedace (Ω. 65 s s s s 6 ero sequece source imedace (Ω. 65 Table II: V distributio cable imedace (Ω / m.65 Number Cable size & tye 5 sqmm Alumiium belted PILCPS [8] Leg L m 3 Positive sequece cable imedace (Ω / m c 4 Negative sequece cable imedace (Ω / c m 5 ero sequece cable imedace (Ω / m cod ero sequece shea resistace (Ω / m. sh 66 7 Effective zero sequece imedace of e g.48 geeral mass of e ear (Ω / m 8 Fault imedace (Ω. Table III: V eutral & earig imedace (Ω umber Neutral imedace (Ω. f oe substatio earig system resistace (Ω [8] MATS 3 MEN imedace of tyical urba extesive MEN system (Ω [8] MEN.3 (MEN is equivalet to me 4 Surface soil resistivity (Ω -m 5 Trasformer earig system resistace (Ω 6 Equivalet MEN lus e imedace (Ω ρ e ρ.346 eq ( MEN e eq..7 Three sequece imedaces ca be foud out by e followig equatios [8]: s ( cod L s ( cod L (8 (9

4 Iteratioal Joural of Scietific ad esearch Publicatios, Volume, Issue, February 4 ISSN ( L ( s cod sh L ( g L 3.( eq MATS ( V. ESULT The fault curret (Phase is tae as iut vector to e wavelet eural etwor. The fault distace is tae as e outut vector of e etwor. Hece e WNN has oe iut ad oe outut dimesio. The etwor has bee traied wi differet (η (µ values of learig rate mometum factor, umber of hidde odes ad total umber of iteratios. A Error versus Iteratio curve is show i Fig 3 which clearly shows at error decreases wi e iteratio. This roves e good covergig roerty of wavelet eural etwor. Error v Iteratio Sum Square Error No of iteratios Figure 3: Sum Square Error value for differet umbers of iteratio The best values of above four arameters are obtaied durig traiig rocess is show i Table IV. Table IV: Best Obtaied Parameters Learig rate Mometum factor No of Hidde Nodes No of Iteratio Usig ese best values of differet arameters, a attemt has bee made to show e accuracy of redictio betwee aalytical ad estimated outut. Sevety six testig data were used for is urose. The fault currets for differet cable locatio are calculated for test cases. Fig 4 shows e grahical reresetatio of aalytically calculated ad estimated values. This shows at aalytical ad estimated values rereseted i solid ad dotted lies resectively. As bo are close eough, which iter shows e caability of e wavelet eural etwor for fault locatio.

5 Iteratioal Joural of Scietific ad esearch Publicatios, Volume, Issue, February 5 ISSN Fault curret v Fault distace Fault curret (A Fault distace (m Aalytical Estimated Fig 4 Aalytical ad estimated values of fault distace VI. CONCLUSION I is aer we desiged bac roagatio algorim based wavelet eural etwor to estimate cable fault (L-G distace. Detailed studies have bee made to determie e effects of various etwor arameters, viz. learig rate (η, mometum factor (µ, umber of wavelos, ad umber of iteratios o covergece of e traiig rocess which was imlemeted by BP algorim. It has bee observed at e WNN has give a accurate redictio of fault distace by usig ese otimum values of etwor arameters. EFEENCES [] H.K. adeh, A Accurate Fault Locator for Cable Trasmissio Usig ANN, Electrotechical Coferece of IEEE Mediterraea, Meleco, Vol. 3,. 9-94, 4. []. Ku, ad C. W. Kai, O-Lie Fuzzy Adative Cotroller wi Wavelet Neural Networ, IEEE Proceedigs, ICII - Beiig, Chia, Vol. 4, ,. [3] A. K. Pradha, A. outray, ad A. Behera, Power Quality Disturbace Emloyig Modular Wavlet Networ, IEEE Coferece o Power Egieerig Society, Malaysia, 6. [4] K. Karieya, A Comarative Study o e Performace of ANN, WNN, ad PSO Aided WNN for Estimatio of Comlex Fuctios i High Voltage System, MEE Thesis, Jadavur Uiversity, Idia, Jue 7. [5] Charu Guta, Imlemetatio of Bac Proagatio Algorim (of eural etwors i VHDL, Master of Egieerig Thesis, Thaar Istitute of Egieerig & Techology (Deemed Uiversity,Patiala, Idia, Jue 6. [6] B.B. Mishra, P.K. Das, ad G. Pada, Patter Classificatio Usig Local Liear Wavelet Neural Networ, IEEE Proceedigs of Iteratioal Coferece o Iformatio ad Commuicatio Techology, Dhaa, Bagladesh,. 9-95, 7. [7].Y. Luo, ad.k. Shi, Wavelet Neural Networ for Fault Diagosis of Push-Pull Circuits, IEEE Proceedigs of Iteratioal Coferece o Machie Learig ad Cyberetics, Guagzhou, Chia, Vol. 6, , August 5. [8] Extesive Urba MEN System Sulied by a HV Buried Cable by htt:// Examle%3.7..%(N.df. [9] J.B. Guta, A Course i Power Systems, by S.K. Kataria & Sos. AUTHOS First Auor Biswariya Chatteree, M.Tech. Assistat rofessor of Deartmet of Electrical Egieerig, Future Istitute of Egieerig & Maagemet, West Begal, Idia Corresodece Auor Biswariya Chatteree, M.Tech. Assistat rofessor of Deartmet of Electrical Egieerig, Future Istitute of Egieerig & Maagemet, West Begal, Idia Phoe No : address : biswaee@gmail.com

Research Article New Topologies of Lossless Grounded Inductor Using OTRA

Research Article New Topologies of Lossless Grounded Inductor Using OTRA Joural of Electrical ad omuter Egieerig Volume 2, Article ID 753, 6 ages doi:.55/2/753 Research Article New Toologies of Lossless Grouded Iductor Usig OTRA Rajeshwari Padey, Neeta Padey, Sajal K. Paul,