COMPLEX NEURAL NETWORK APPROACH TO OPTIMAL LOCATION OF FACTS DEVICES FOR TRANSFER CAPABILITY ENHANCEMENT

Transcription

1 ARPN Journal of Engneerng and Appled Scences Asan Research Publshng Networ (ARPN). All rghts reserved. COMPLEX NEURAL NETWORK APPROACH TO OPTIMAL LOCATION OF FACTS DEVICES FOR TRANSFER CAPABILITY ENHANCEMENT D. Venu Madhava Chary 1 and J. Amarnath 1 Department of Electrcal and Electroncs Engneerng, M V S R Engneerng College, Hyderabad, A.P., Inda Department of Electrcal and Electroncs Engneerng, J N T U H College of Engneerng, Hyderabad, A.P., Inda E-Mal: dvm_chary@yahoo.com ABSTRACT Ths paper concentrates on enhancement of total transfer capablty ncorporatng FACTS devces. Repeated power flow program s used to determne the voltage constraned total transfer capablty (TTC). The effect of change n reactance of the lne on the transfer capablty and reactve power loss s studed. In recent years Complex valued Artfcal Neural Networs (ANNs) are becomng popular for solvng problems nvolvng complex data. As the transfer capablty, real and reactve power loss depends on the lne parameters a novel method for computng transfer capablty and total real and reactve power loss s proposed n ths paper usng complex valued neural networ. Keywords: FACTS devces, TCSC, transfer capablty enhancement, complex valued neural networs. INTRODUCTION Power system load growth s ncreasng at a faster rate as compared to the ncrease n transmsson capablty. In the last decade the ncrease n transmsson capacty s approxmately 50% of the ncreased generaton capacty. The varous reasons for the short fall are dffculty n gettng rght of way and permts due to property devaluaton, electromagnetc feld effects on health, mpact on land use, ecologcal systems and constructon and mantenance. The other mportant reason s lac of nvestors for the proposed transmsson proects. All the above factors and the rapd growth of the load lead to congeston of lnes. The performance of the system can be mproved wthout reschedulng or topologcal changes by ncorporatng FACTS devces [1-]. Several methods are proposed for ths purpose. Wu G., Yooyama A., He J., Yu Y., suggested the allocaton and control of FACTS devces for stablty mprovement [3]. Preedavncht P, Srvastava S.C., presented an approach for reactve power dspatch [4]. Optmal flow wth FACTS devces s dscussed n references [5, 6]. In reference [7] economc dspatch method s used for fndng out the optmal locaton of FACTS devce. As seres compensaton s most effectve way of enhancng power transfer capablty n ths paper a method to determne the optmal locaton of Thyrstor Controlled Seres Capactors (TCSC) usng complex valued neural networ s proposed. Ths approach s based on determnng the actual lne transfer capablty (real and reactve flows) and total power loss (real and reactve power loss) by ntroducng TCSC n each lne for a voltage constraned power system. TRANSFER CAPABILITY, CONSTRAINTS For plannng and operaton of a power system determnaton of transfer capablty s very mportant. For a secured system transfers should not exceed the transfer capablty lmts. One of the most common approaches for transfer capablty calculatons s the contnuaton power flow (CPF) [8, 9]. Power transfer capablty of a power system ndcates that how much nter area power transfer can be ncreased wthout any securty volatons [10]. For a bul power system exact computaton of the transfer capablty provdes vtal nformaton for plannng and secured operaton. The man constrants whch lmt the transfer capablty are: Thermal related Voltage related and Operaton related. Thermal related constrants may cause overheatng of conductor thereby decreasng ts lfe span. The lne sag also ncreases wth temperature whch may cause conductor to ground clearance volaton and loss of mechancal strength. Voltage related constrants cause the voltage dp or swell problems whch may n turn cause nsulaton falure, nterference wth adacent communcaton lnes, nadequate operaton and damage of equpment. FACTS DEVICES Wth the advent of flexble ac transmsson system (FACTS) devces power utltes all over the world are able to mprove the system stablty lmt, control the power flow, mprove the transmsson system securty and provde strategc benefts for better utlzaton of the exstng power system. The operaton of FACTS devces s based on power electronc controllers. These devces are also used to enhance transfer capablty and to mnmze the total power loss of a system thereby mprovng the system effcency. In a compettve electrc power system the most mportant aspect s better utlzaton of exstng lnes n the context of growng demand and outgrowth of energy tradng marets. In the context of restructurng the exstng power systems FACTS devces have assumed an mportance snce they can expand the usage potental of transmsson systems by controllng power flows n the networ. FACTS devces are operated n a manner so as to 1

2 ARPN Journal of Engneerng and Appled Scences Asan Research Publshng Networ (ARPN). All rghts reserved. ensure that the contractual requrements are fulflled as far as possble by mnmzng lne congeston. The concept of FACTS devces was frst proposed by Hngoran [11]. The power system can be operated n a more secured, flexble and sophstcated way wth the help of FACTS devces. In present deregulated envronment the operaton of power system need much more sophstcated means of power control whch can be met by the ncorporaton of FACTS devces. STATIC MODELING OF TCSC Thyrstor controlled seres capactors (TCSC) are connected n seres wth transmsson lnes. It s equvalent to a controllable reactance nserted n a lne to compensate the effect of the lne nductance. The net transfer reactance s reduced and leads to an ncrease n power transfer capablty. The voltage profle as also mproved due to the nserton of seres capactance n the lne. The transmsson lne model wth a TCSC connected between the two buses and s shown n Fgure-1. Equvalent p model s used to represent the transmsson lne. TCSC can be consdered as a statc reactance of magntude equvalent to -X c. The controllable reactance Xc s drectly used as control varable to be mplemented n power flow equaton. Fgure-1. Transmsson lne model wth TCSC. The followng equatons are used to model TCSC. Let the voltages at bus and bus are represented by V δ and V δ The complex power from bus to s S P Q V I (1) [( V V ) Y V ( Bc )] [ G + ( B + B )] V V ( G B )] V + () V + (3) Where G c + B 1 (4) ( R + X X ) L L From the above equatons the real and reactve power equatons can be wrtten as P V G V V G C cos( δ δ ) Q VV B sn( δ δ ) (5) V ( B c + B ) V V G sn( δ δ ) + V V B cos( δ δ ) (6) Smlarly the real and reactve powers from bus to can also be represented replacng V by V. The real and reactve power losses n a lne are represented by equatons (7) and (8). P L P + P (7) Q L Q + Q (8) LOCATION OF FACTS DEVICES The obectves for devce placement may be one of the followng: a) Reducton n the real power loss of a partcular lne b) Reducton n the total system real power loss c) Reducton n the total system reactve power loss d) Maxmum relef of congeston n the system For the frst three obectves, methods based on the senstvty approach may be used. If the obectve of FACTS devce placement s to provde maxmum relef of congeston, the devces may be placed n the most congested lnes or, alternatvely, n locatons determned by tral-and-error. COMPLEX VALUED NEURAL NETWORK In recent years, complex-valued neural networs have wdened the scope of applcaton to the problems where the nput and output are n complex form such as load flow analyss, contngency analyss. As nputs to CVNNs are n complex form generalzaton algorthms are modfed to manpulate complex valued data. Complex bac propagaton algorthm can be appled to multlayered neural networs whose weghts, threshold values, nputs and outputs all are complex numbers. Complex verson of bac propagaton (CVBP) algorthm made ts frst appearance when Wdrow, Mc Cool and Ball [1] announced ther complex least mean squares (LMS) algorthm. Hrose [13] studed the dynamcs of CVNN whch was later appled to the problem of reconstructng vectors. An extensve study of CVBP was reported by Ntta [14]. L. Chan, A.T.P. So and L.L. La [15] publshed the frst paper on applcatons of complex artfcal neural networs to load flow analyss. Complex valued neural networ requres half the number of nputs used by the conventonal real valued neural networ. Furthermore the average learnng speed of complex bac propagaton algorthm s several tmes faster than that of real valued bac propagaton.

3 ARPN Journal of Engneerng and Appled Scences Asan Research Publshng Networ (ARPN). All rghts reserved. Voltage constraned total transfer capablty and total power loss are obtaned by repeated power flow method. Fgure-. Complex valued neural networ. Internal potental of hdden neuron : u I 1 ( w x ) + θ Re[ u ] + Im[ u ] (9) Output of hdden neuron : z φ ( u ) + Re[ z ] Im[ ] Re[ u ] Im[ u ] + z (10) + e + e Internal potental of output neuron : s m 1 ( v z ) + γ Re[ s ] + Im[ s ] (11) Output of output neuron : o φ ( s ) + Re[ y ] Im[ ] Re[ s ] Im[ s ] + y (1) + e + e Error E 0.5 l 1 (Re[ y ] Re[ d ]) + (Im[ y ] Im[ d ]) (13) Wth the help of ths error E we derve the gradent of E wth respect to both the real and magnary part of the complex weghts of nput layer and output layer E/ w and E/ v respectvely. Durng tranng the networ cost functon E s mnmzed by recursvely alterng the weght coeffcent based on gradent descent algorthm, gven by equaton (15), w ( p + 1) w ( p) + w ( p) w ( p) η E (15) w p w w Where p s the number of teratons and η s the learnng rate constant. RESULTS AND DISCUSSIONS The proposed method s appled to a 14 bus system shown n Fgure-3. It conssts of 0 lnes and 5 generators. Only 15 lnes are consdered for locaton of TCSC. The effect of TCSC n a partcular lne s consdered as change n lne reactance component. ( p ) Fgure Bus system. Tranng patterns for the proposed complex valued neural networ are obtaned by varyng the lne reactance from 0% to 80 % n steps of random values and the effect on each lne s studed. The networ taes the lne admttance n complex form as nput; the outputs are lne flow and total power loss n complex form. Complex bac propagaton algorthm s used to tran the networ. The number of teratons requred s 5000 wth an error of The varaton of error wth respect to number of teratons s shown n Fgure-4. Error Iteratons Fgure-4. Varaton of error wth number of teratons. Fgure-5 represents the effect of controllng lne reactance ncorporatng TCSC on total reactve power loss of the system. 3

4 ARPN Journal of Engneerng and Appled Scences Asan Research Publshng Networ (ARPN). All rghts reserved. Reactve power loss Transfer Capablty Transmsson Lnes Fgure-5. Reactve power loss wth TCSC n each lne Transmsson Lnes Fgure-6. Transfer capablty varatons of each lne. The proposed networ s tested for dfferent values of lne reactance ncorporatng the effect of TCSC as shown n Tables [1 to 4]. Table-1. Transfer capablty of lne 1-5 and total loss. Lne 1-5 wth 55% reactance Desred output Lne flow Total loss Table-. Transfer capablty of lne -3 and total loss. Lne -3 wth 70% reactance Desred output Lne flow Total loss Table-3. Transfer capablty of lne -4 and total loss. Lne -4 wth 55% reactance Desred output Lne flow Total loss Table-4. Transfer capablty of lne -5 and total loss. Lne -5 wth 55% reactance Desred output Lne flow Total loss It s observed that there s a neglgble change n reactve power loss at lnes 9, 11 and 15. The mnmum reactve power loss s 396 MVAR wth 0% reactance on lne 1-5 where as the base case reactve power loss s 58MVAR. The varaton of transfer capablty of the lnes wth change n reactance due to the ncorporaton of TCSC s shown n Fgure-6. It s observed that the enhancement of transfer capablty of lnes 9, 11, 1 and 14 s very small. Consderng the reactve power loss the optmal locaton of TCSC s n lne 1-5. REFERENCES [1] Galana G.D Assessment and control of the mpact of FACTS devces on power system performance. IEEE Transactons on Power System. 11(4): [] Larsen E., Mllers N., Nlsson S., Lndgren S Benefts of GTO-Based compensaton systems for electrc utlty applcatons. IEEE Transactons on Power Delvery. 7(4): [3] Wu G., Yooyama A., He J., Yu Y Allocaton and control of FACTS devces for steady state stablty enhancement of large scale power system. IEEE Internatonal Conference on Power System Technology. 1: [4] Preedavncht P., Srvastava S.C Optmal reactve power dspatch consderng FACTS devces. Electrc Power Systems Research. 46(3): [5] Momoh J.A., Zhu J.Z A new approach to optmal power flow wth phase shfter. IEEE Internatonal Conference on Systems. 5: [6] Le T.T., Deng W Optmal flexble AC transmsson systems (FACTS) devces allocaton. Internatonal Journal of Electrcal Power and Energy Systems. 19():

5 ARPN Journal of Engneerng and Appled Scences Asan Research Publshng Networ (ARPN). All rghts reserved. [7] De Olvera E.J., Lma W.M Allocaton of FACTS devces n a compettve envronment. 13 th PSCC. pp [8] V. Aarapu and C. Chrty The Contnuaton Power Flow: A tool for steady state Voltage stablty Analyss. IEEE Transactons on power systems. 7(1): [9] H.D. Chnag, A.J. Fluec, K.S. Shah and N. Balu CP FLOW: A practcal Tool for Tracng Power system Steady state statonary behavour due to the load and generaton varatons. IEEE Transactons on power systems. 10(): [10] Transmsson Transfer Capablty Tas Force (NERC). Avalable transfer capablty defntons and determnaton. [11] Hngoran N.G, Gyugy L Understandng FACTS: Concepts and Technology of Flexble AC Transmsson Systems. Insttute of Electrcal and Electronc Engneers Press, New Yor. [1] B. Wdrow, J. Mc Cool and M. Ball The Complex LMS algorthm. Proc. of the IEEE. Aprl. [13] A. Hrose. Dynamcs of fully complex-valued neural networs. Electroncs letters. 8(13): [14] T. Ntta An extenson of the bac- propagaton algorthm to complex numbers. Neural Networs. 10(8): [15] L. Chan, A.T.P. So and L.L. La Intal applcatons of complex artfcal neural networs to load flow analyss. IEE Proc. Gener. Transm. Dstrb. 147(6): , November. 5