International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:03 7

Transcription

1 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 7 Applying Muliple Paricle Swarm Opimizaion Algorihm o he Opimal Seing of Time Coordinaion Curve of in Disribuion Feeder Auomaed Sysem Ming-Yuan Cho, Member, IEEE, Shen-Wen Hsiao, Chia-Yuan Hsu Absrac In his sudy, mahemaical opimizaion was applied o improve he faul inerval assessmen of feeder erminal unis (s). Because of he insufficien hardware compuing power in such unis, using a ime coordinaion curve o calculae faul curves leads o he occurrence of jagged curves. Such jagged curves lead o considerably high errors, ulimaely resuling in malfuncion or erroneous nonoperaion. A muli-paricle swarm opimizaion (muli-pso) algorihm was proposed for resolving he malfuncion caused by proecion curves wih considerably high errors. This algorihm improves he early convergence problem encounered in he original sandard PSO algorihm and he long calculaion ime observed in he area mehod. When he proposed muli-pso algorihm is applied in he opimizaion process, he arrangemen of he nodes on he jagged curves improved; his resuls in he jagged curves exhibiing relaively smooh and curved oulines, hereby eliminaing malfuncions and achieving upsream and downsream proecion coordinaion as well as compleing faul inerval idenificaion procedures. The main conribuion of his sudy is he improvemen of he erroneous assessmen of disribuion line fauls observed in faul flags. The proposed mehod also improves he success rae of FDIR in feeder auomaion sysems. Index Term Faul diagnosis isolaion resoraion (FDIR), Feeder auomaion, Feeder erminal uni (), Muli-paricle swarm algorihm, Proecion coordinaion. I. INTRODUCTION TO improve he power supply qualiy and shoren he ouage ime, he Taiwan Power Company (Taipower) has esablished feeder auomaion sysems on is disribuion nework. During persisen fauls in he disribuion sysem, he exper sysem for faul diagnosis isolaion resoraion (FDIR) wihin he feeder auomaion sysems deermines he faul sie and isolaes he inerval of he inciden according o he faul flag generaed by he feeder erminal uni (). The load of he disribuion sysem is hen analyzed so ha i can be used o resore power upsream and ransfer power downsream while mainaining adequae supply qualiy. This shorens he ouage inerval and reduces user ouage ime, improving he operaional efficiency of feeder auomaion sysems [1]. faul flagging involves using a ime coordinaion curve (TCC) for deecing and assessing fauls. A key influence on wheher an can flag fauls appropriaely is is mehod of calculaing he TCC. Because hardware has insufficien compuing capabiliy, TCC faul curves are no exhibied as consecuive poins arranged in an arc shape. Raher, nodes on a jagged curve are used o deec faul curren. However, using jagged curves for deecing faul curren ofen resuls in considerably large errors, prevening upsream-downsream proecion coordinaion. Thus, when an deecs faul curren, abnormaliies such as excessive flagging or failure o flag occur as shown in Fig. 1. Failure o flag ofen leads o equipmen meldown, whereas excessive flagging usually causes power ouages. Inadequae faul flagging accuracy also prevens he FDIR exper sysem in he feeder auomaion sysem from execuing is proper funcion when feeder incidens occur. Thus, he sysem or he dispacher misjudges he inciden inerval, which expands he inciden range and severely affecs he qualiy of he power supplied o users. Numerous sudies have applied paricle swarm opimizaion (PSO) o seek opimal soluions; for example, PSO algorihms have been applied o power flow problems involving disribued generaor failures, o he design of power sysem sabilizers, and o he deerminaion of opimal posiions for deploying volage measuremen equipmen in power sysem planning [2] [4]. However, few sudies have applied PSO o solve TCC opimizaion during fauls. Taipower currenly uses a rule-of-humb approach hrough human-machine inerfaces o address he opimizaion of faul curves. This approach usually resuls in malfuncions induced by excessive flagging or failure o flag [5]. In addiion, mos sudies have solved and compared he opimizaion of correlaion curves, such as linearly decreasing curves, exponenial curves, parabolic curves opening upward, fuzzy curves, and random curves, by using various PSO algorihms [8], [9]. Afer feeder fauls are diagnosed, he faul posiion mus be confirmed and isolaed. A his ime, an accurae TCC mus be se o isolae he faul segmen appropriaely. The curve ype in his siuaion is primarily an inverse ime-curren coordinaion curve. Alhough PSO can be used o obain he soluion for his ype of curve in a relaively quick manner, i resuls in premaure convergence and local opimal soluions [9]. In he curren sudy, a muli-pso algorihm was applied o derive opimal soluions. In he proposed mehod, a secondary sandard paricle swarm

2 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 8 algorihm is execued o derive global opimal soluions while invesigaing opimal TCC se poins. Acion ime Acion ime Curve obained when he fuse sared meling Curve obained when he fuse meled compleely Acual faul curve Ideal faul curve faul curve Failure o flag Faul curren (a) Excessive flagging. Curve obained when he fuse sared meling Curve obained when he fuse meled compleely Acual faul curve Ideal faul curve faul curve Excessive flagging Faul curren (b) Failure o flag. Fig. 1. Excessive flagging and failure o flag for faul deecion. II. FDIR EXPERT SYSTEM IN THE FEEDER AUTOMATION OF DISTRIBUTION SYSTEM To alleviae ouage incidens, Taipower has repaired feeders manually in recen years. During such repair processes, field personnel ofen experience accidens, repair imes are excessively long, and users are inconvenienced. To ameliorae hese problems, Taipower is currenly inroducing an FDIR exper sysem ino is feeder auomaion promoion sysem, he Siemens SINAUT Specrum 4.4. This sysem is projeced o increase power qualiy [6], shoren he ouage ime, and reduce accidens experienced by field personnel. The FDIR exper sysem uses supervisory conrol and daa acquisiion (SCADA) o opimize feeders o upgrade he daabase. Disribuion feeders having environmenal geographic maps as blue plaes are added o he SCADA daabase wihin he feeder opimizaion sysem. This faciliaes he process of monioring acual disribuion feeders because he maps of such disribuion feeders are available wihin he sysem daabase. The calculaes nodes o deermine he faul curve. When he curve of he faul curren is consisen wih he faul curren curve deermined by he, he reurns a faul flag o he maser saion, as shown in Fig. 2-(a). An environmenal geographic map among he SCADA funcions is hen used o judge he locaion of he inciden, as shown in Fig. 2-(b). Nex, an auomaic command insrucs he o execue he auomaic line swich o isolae he inciden segmen, as shown in Fig. 2-(c). Afer he inciden segmen is isolaed, he sysem auomaically sends a conrol command o he remoe erminal uni of he subsaion o inpu he feeder circui breaker (). This enables he sound segmen upsream of he inciden segmen o resore power. Finally, he load of he disribuion sysem downsream is analyzed by using a ransfer feeder ha in principle has he same alernaing volage and subsaion as ha of he disribuion sysem as well as having a sufficien load capaciy. This enables he disribuion sysem o recommend a ransfer plan for he resoraion of downsream power while mainaining adequae power qualiy. The dispachers on duy a he Taipower feeder dispach conrol cener reference his plan, enabling power o be resored in sound downsream segmens, as shown in Fig. 2-(d). This shorens he ouage inerval, reduces user ouage ime, and improves he operaional efficiency of feeder auomaion. Conrol cenre Conrol cenre Conrol cenre Conrol cenre A Subsaion B Subsaion Faul flag Faul flag (a) Faul flag is generaed (F). A Subsaion Trips B Subsaion Faul flag Inciden Inciden inerval Faul flag (b) Faul is deeced auomaically (FD). A Subsaion Trips B Subsaion Faul flag Inciden inerval Incision Faul flag (c) Faul is isolaed auomaically (I). A Subsaion Inpu B Subsaion Faul flag Incision Inciden inerval Incision Faul flag Incision (d) Upsream and downsream power is resored auomaically (R). Fig. 2. Schemaic of he operaions of he FDIR exper program. Inpu

3 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 9 III. PROBLEM FORMULATION A. Planning of he Faul Curve in he Feeder Auomaion Sysem The faul curve for he Taipower feeder auomaion sysem is currenly formulaed based on he faul curve equaion of he inelligen elecronic device (IED) embedded in he of each subsaion. In (1), he faul curve of he REF-541 IED relay produced by ABB is used o formulae he faul curve. According o Taipower s secondary subsaion proecion coordinaion provisions and Briish Sandards 142 or Inernaional Elecro echnical Commission (IEC) sandard, parameers associaed wih he exremely inverse curve ype shown in Table 1 are used for formulaing he faul curve [] [12]. (1) In his formula, : Acion ime (s). : Adjusable ime muliplier (his can generally be considered as he lever; ha is, he se ime value). : Phase curren (A) (his can be considered as he acual curren or he es curren). : Adjusable curren pick-up value (A) (his can generally be considered as he ap; ha is, he se curren value). and : consans. Their values indicae he curve ype used (Table I). Table I Parameers for he IED proecion curve ypes. Consans Normal Inverse Highly Inverse Exremely Inverse Mos of he proecion relays in s have been modified from radiional mechanical proecion relays o IEDs. The IED es manual indicaes ha he error value of he acion ime for he pick-up curren value,, is wihin ±% [7] and he error value of he hardware response ime of he faul curve is wihin ±%. To preven he error value of he coordinaion ime from affecing he proecion coordinaion beween he and curves, he maximum error value of he proecion ime of he wo curves mus be 20% (Fig. 3). This value is used o se he faul curve by using (2), achieving he provisions for upsream and downsream proecion coordinaion. This error parameer can also be adjused manually. Acion ime Curve obained when he fuse sared meling (primary proecion) Curve obained when he fuse meled compleely (primary proecion) faul curve faul curve (backup proecion) Coordinaion ime >20% Faul curren Fig. 3. faul curve planning posiions. B. Finess Funcion The finess funcion is used in opimizaion processes for deermining he figure of meri. In his sudy, he jagged area beween he ideal faul curve and he acual faul curve were used o deermine he figure of meri. Fig. 4 shows ha he faul curve is enhanced as he jagged area diminished. As shown in Fig. 4, poins a, b, c, and d on he faul curve are hardware nodes and correspond o he size of he faul curren on he x-axis. To obain he opimized faul curve, fixed values mus be se for poins a and d on he curve. The finess funcions of poins b and c are hen obained o deermine he figure of meri. Acion ime a b Curve obained when he fuse sared meling Curve obained when he fuse meled compleely Acual faul curve Ideal faul curve faul curve c Faul curren Fig. 4. Finess Funcion. To obain he finess funcion of poin b in he ih paricle swarm, poins a and c mus be se o fixed values. The inegrals of he area under he curves formed by poins a b and poins b c are hen calculaed. The inegral of he hached area under he curve formed by poins a b is calculaed using (3), whereas ha of he curve formed by poins b c is calculaed using (4). In hese formulae, = 1, 2, 3,, m is he number of ieraions, i = 1, 2, 3, and n is he paricle swarm [9]. d (2) (3) (4)

4 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 The inegral areas of (3) and (4) are hen added as shown in (5). As he area of decreases, he finess funcion of poin b in he ih paricle swarm becomes enhanced. (5) To deermine he finess funcion of poin c in he ih paricle swarm, poins b and d mus firs be se o fixed values. The inegral of he shaded area under he curve formed by poins b c is calculaed using (4), whereas ha of he curve formed by poins c d is calculaed using (6). (6) The inegral areas of (4) and (6) are hen added as shown in (7). As he area of decreases, he finess funcion of poin b in he ih paricle swarm becomes enhanced. (7) In he overall finess funcion of he faul curve, he enire jagged area formed by he ideal faul curve and acual faul curve is used o deermine he figure of meri of he hardware nodes, as shown in (8). (8) The problem in his sudy can be solved using he opimized mahemaical model shown in (9). (9) Subjec o In his formula, Consrain 1 indicaes ha poins b and c mus be locaed wihin he inerval of he fuse complee meling curve and he faul curve. Consrain 2 saes ha he ime inerval beween he faul curve and faul curve mus be a leas 20% of he pu ime value. IV. DEVELOPMENT OF METHODOLOGY A. Sandard Paricle Swarm Algorihm Three facors generally influence he direcion of a paricle a he nex poin in ime: paricle direcion a he nex poin in ime = paricle direcion a ime + opimal direcion deermined by he paricle + opimal direcion deermined by he swarm. In 1998, Shi and Eberhar proposed an ineria weigh w in heir implemenaion of he sandard PSO algorihm shown in (). Various weigh calculaions can be used o updae he convergence speed auomaically. The weigh is generally iniialized a 0.9 and declines linearly o 0.4. () where : Ineria weigh. : Speed of paricle i a ime. : vecor of paricle i a ime. : Opimal posiion deermined by paricle i so far. : Opimal posiion deermined by he paricle swarm so far. : Learning facor, also called he acceleraion facor_weigh of individual experience, ypically se o 2. : Learning facor, also called he acceleraion facor_weigh of swarm experience, ypically se o 2. : Random value wihin [0, 1]. : Random value wihin [0, 1]. When he ineria weigh, w, is relaively high, he PSO algorihm can search for a wide range of values. When he ineria weigh is relaively low, he PSO algorihm can search for a narrow range of values. Therefore, if a linearly decreasing ineria weigh is used during he ieraive calculaion, such as ha shown in (11), he PSO algorihm can exhibi excellen global search capabiliies from he beginning and can quickly locae a region ha is near he global opimal soluion. In he laer sages, i can also exhibi excellen local search capabiliies and can accuraely deermine he global opimal soluion. where : Iniial ineria weigh. : Final ineria weigh. : Maximum number of ieraions. : Curren number of ieraions. (11) To use he sandard PSO algorihm o obain he opimised faul curve, mus be redefined. Each noe on he faul curve represens an posiion, and each posiion has i paricle swarms o obain he opimal soluion. The algorihm shown in () mus be redefined as (12) and (13): where (12) (13) : Speed of he ih paricle swarm a ime on he node. : vecor of he ih paricle swarm a ime on he node. : Opimal posiion deermined by he enire swarms g before ime on he node. : Opimal posiion deermined by he ih swarm before

5 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 11 ime on he node. When paricles are searching for opimizaion, he maximum speed and minimum speed resric he speed of. In addiion, influences a paricle s search capabiliies in he region beween is curren posiion and is arge posiion. A considerably high induces paricles o fall ou of a region wih a srong soluion. Conversely, a considerably low induces paricles fall ino a local opimal soluion [9]. B. Muli-PSO Muli-PSO is an advanced swarm opimizaion algorihm ha is based on he sandard PSO algorihm. I is used o obain a coninuous muli-arge soluion [8], [9]. The calculaion principle of muli-pso is o simulaneously execue a sandard PSO wice o obain he opimal soluion. The firs execuion of he sandard PSO algorihm is conduced o deermine muliple arges hrough a global search o shrink he opimal soluion range for each arge. The second execuion is conduced sequenially for each arge while shrinking he opimal soluion range o generae opimal soluions for single arges. Local convergence of single arges is used o influence he oher arge soluions and hereby obain he overall opimal soluion. When he sandard PSO algorihm is used o obain he opimized faul curve, node b ofen influences node c, forcing node c ino a local opimizaion soluion. Thus, muli-pso is used for calculaion o improve his problem. In muli-pso, he firs execuion of he sandard PSO algorihm is conduced using (7) and (8) o deermine he opimal region on he faul curve in which a global search for he nodes can be conduced. This enables each node o move o he opimal region on he curve, as shown in Fig. 5-(a). During his global search, he sandard PSO algorihm is ieraed sequenially on each node beween he firs and final node on he faul curve o obain a local convergence of he curve, as shown in Fig. 5-(b). This improves he problem of he sandard PSO algorihm described in his paper. Acion ime a b Curve obained when he fuse sared meling Curve obained when he fuse meled compleely Acual faul curve Ideal faul curve faul curve Opimal region of node (b) Opimal region of node (c) c d Faul curren (a) Global search. Acion ime a b Curve obained when he fuse sared meling Curve obained when he fuse meled compleely Acual faul curve Ideal faul curve faul curve Opimal region of node (b) Opimal region of node (c) c d Faul curren (b) Local convergence. Fig. 5. Muli-PSO algorihm. When he second sandard PSO algorihm is execued sequenially for each node, he finess funcion is se o an unopimized value, as shown in (14). During he nex ieraion, is se o he iniial speed, as shown in (15) and (16). This causes he node o remain in is curren posiion on he curve. However, his resuls in an unopimized finess funcion and a speed in he nex ieraion ha is sufficien o exi local soluions. (14) (15) In hese formulae, : Nodes. : Number of ieraions. : Paricle swarm. (16) Afer he muli-pso algorihm is ieraed again, he curren finess funcion of he single node becomes more opimised han i was before, as shown in (17). Formula (16) is hen used o produce a new o exi he curren local soluion. Nex, (12) is used o generae new and values o produce he nex new value. This value is used o coninue he opimizaion of he overall faul curve. C. Muli-PSO Calculaion Process (17) The proposed muli-pso algorihm is based on he sandard PSO algorihm. I improves he limiaions of he sandard PSO algorihm, which ofen falls ino local soluions when obaining he opimizaion of faul curves. The basic seps of he muli-pso algorihm are oulined as follows: Sep 1: The parameer seings are iniialized. The maximum and minimum speed of each paricle, maximum and minimum posiion vecor of he faul curve, number of paricle swarms, number of ieraions, ineria weigh, and learning facors ( and ) are se.

6 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 12 Sep 2: The speed of each paricle and posiion of he faul curve are iniialized randomly. Sep 3: The finess value of each paricle on every posiion is calculaed. Sep 4: The finess value of each paricle on every posiion is assessed. The posiion for he finess value of each paricle on every posiion is se as, he opimal posiion for each individual paricle. Sep 5: The opimal posiion for each individual paricle on every posiion is again assessed. The mos favourable posiion among he resuling opimal posiions is se as, he opimal posiion for he paricle swarm. Sep 6: This value is subsiued ino (12) and (13) o updae he speed and posiion of each paricle. Sep 7: Confirm wheher he number of ieraions has reached he sop condiion. YES Oupu he opimal soluion. NO Proceed o Sep 8. Sep 8: Confirm wheher he number of ieraions has reached he improvemen plan condiion. YES Proceed o Sep 9. NO Reurn o Sep 3. Sep 9: For each ime he improvemen plan condiion is me, he finess value of each paricle on every posiion is subsiued sequenially ino (14) and is subsiued ino (15). Reurn o Sep 3. Fig. 6 shows he overall flowchar of hese seps. NO Sar Iniialise parameer seings: Maximum speed of each paricle v max Minimum speed of each paricle v max Maximum posiion of he faul curve Minimum posiion of he faul curve Number of paricle swarms i Number of ieraions G max Ineria weighw Learning facor 1 c c 2 Randomly iniialise: Speed of each paricle v faul curve posiion node _ i x node _ i x x Anode _ i max Calculae he finess funcion for he posiion of each paricle on all posiions Updae based on he finess value: Opimal posiion for each individual paricle Opimal posiion for he paricle swarm min p p node _ Updae he speed and posiion of each paricle based on he posiion and speed formulae Has he number of ieraions reached he end condiion? NO Has he number of ieraions reached he improved plan condiion? YES g node _ i YES End 1 v Updae he speed node _ i and finess value _ of each paricle based on he muli-pso algorihm Anode i Fig. 6. Process of solving he opimizaion of he faul curve. V. SIMULATION RESULTS AND DISCUSSION A. Parameer Seing for he Muli-PSO Algorihm We compared he advanages and disadvanages of using he muli-pso and sandard PSO algorihms o opimize he faul curve. The comparison was conduced using he resuls calculaed hrough he minimum area mehod. Table 2 shows he relevan iniialized parameer seings. Excep for he number of muli-paricle ieraions, which was assigned only o he muli-pso algorihm (i.e., 20 ieraions), he same parameers were used in he wo algorihms. The speed of each paricle was iniialized randomly. Regarding curve seings, he faul curve was se according o he exremely inverse curve of he REF-541 IED manufacured by ABB. Curve opimizaion was conduced on he same curve according o he IEC sandard.

7 S S Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 13 Parameer Seings Table II Iniialized parameer seings. Algorihm Muli-PSO Algorihm Faul Curve Sandard PSO Algorihm Maximum Speed of Each Paricle Minimum Speed of Each Paricle 5 5 Maximum Vecor of he Faul Curve 1, 1, Minimum Vecor of he Faul Curve Number of Paricle Swarms 3 3 Number of Ieraions 1,000 1,000 Ineria Weigh Learning Facors, Number of Muli-Paricle Ieraions 20 0 The primary objecive of his sudy was o improve he jagged oulines of he curve. Table 3 shows he simulaion resuls of he node posiions on he curve. The lowes posiion, highes posiion, and middle posiion of he curve for nodes b and c were se; nodes b and c were placed a he lowes and highes posiions separaely. The wo algorihms and he minimum area mehod were used o compare he oal jagged area. The advanages and disadvanages of each mehod were hen analyzed. Table III s of nodes b and c. Curve Number Iniial of Iniial of Node c 1 2 1, 3 1, 4 1, B. Simulaion Resuls and Discussion Figs. 7-(a) and 7-(c) show he rajecory diagram obained when muli-pso was used o calculae he faul curve of posiions 1, 4 (Table 3) for nodes b and c. Figs. 7-(b) and 7-(d) show he changes in he oal jagged area obained when muli-pso was used o calculae he faul curve of posiions 1, 4 in (Table 3) for nodes b and c. Minimum posiion Acion ime MPSO (Modified Paricle Swarm Opimisaion) faul curve Faul curren(a) faul curve Ideal faul curve Acual faul curve Maximum posiion oal jagged area MPSO oal jagged area Number of Ieraions (a) Trajecory diagram of he faul (b) Changes in he oal jagged area curve of nodes b (a ) and of he faul curve of nodes c (a ). b (a ) and c (a ). Paricle1 Paricle2 Paricle3 Minimum posiion Acion ime MPSO (Modified Paricle Swarm Opimisaion) faul curve Faul curren(a) faul curve Ideal faul curve Acual faul curve Maximum posiion oal jagged area MPSO oal jagged area Number of Ieraions (c) Trajecory diagram of he faul (d) Changes in he oal jagged area curve of nodes b (a ) and of he faul curve of nodes c (a 1). b (a ) and c (a 1). Fig. 7. Trajecory diagram and changes in he oal jagged area of he muli-pso faul curve. Fig. 8-(a) and 8-(c) show he rajecory diagram obained when muli-pso was used o calculae he faul curve of posiions 1, 4 (Table 3) for node b. Fig. 8-(b) and 8-(d) show he rajecory diagram obained when muli-pso was used o calculae he faul curve of posiions 1, 4 (Table 3) for node c. Node Node MPSO nodes b Number of Ieraions Paricle1 Paricle2 Paricle3 Node MPSO nodes c Number of Ieraions (a) Calculaed rajecory diagram (b) Calculaed rajecory diagram of node b (a ). of node c (a ). MPSO nodes b Number of Ieraions Paricle1 Paricle2 Paricle3 Node MPSO nodes c Number of Ieraions (c) Calculaed rajecory diagram (d) Calculaed rajecory diagram of node b (a ). of node c (a 1). Fig. 8. Muli-PSO-calculaed rajecory diagrams of nodes b and c. Table 4 shows ha during he opimizaion of he faul curve by using he sandard PSO algorihm, he iniial posiions of nodes b and c are and ; his able indicaes ha afer he ieraions, he posiions of nodes c and b obained using he minimum area mehod are close o heir iniial posiions. The oher iniial posiions are considerably differen from he posiions of nodes c and b obained using he minimum area mehod. The reason for his is ha node b influences node c when he sandard PSO algorihm is used o opimize he faul curve. The origin of node c moves lef and righ, leading o he ieraion resuls of he node c posiion being virually he same as hose of he iniial node c posiion. Thus, during he execuion of he sandard PSO algorihm, node c falls ino local soluions, resuling in early convergence. This affecs he opimizaion of he overall faul curve. To improve he early convergence caused when sandard PSO falls ino local soluions, he muli-pso algorihm is proposed. Table IV shows ha he node posiions in he ieraion resuls Paricle1 Paricle2 Paricle3 Paricle1 Paricle2 Paricle3 Paricle1 Paricle2 Paricle3

8 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 14 obained using he muli-pso algorihm on any posiion on he faul curve are close o he opimal node posiions obained using he minimum area mehod. Figs. 7 and 8 indicae ha his solves he limiaion of he sandard PSO, which is likely o fall ino local soluions and cause early convergence. Table IV Node posiions. Paricle Swarm Paricle 1 Paricle 2 Paricle 3 Paricle 1 Paricle 2 Paricle 3 Paricle 1 Paricle 2 Paricle 3 Paricle 1 Paricle 2 Paricle 3 1, 1, 1, Iniial Seing Sandard PSO Muli-PSO Node c 1, 1, 1, 1, 1, 1, Node c , , , , , , , , , Node c Minimum Area Mehod Node c Table 5 shows ha afer he various node posiions were calculaed, he oal jagged area obained using muli-pso is considerably smaller han ha obained using sandard PSO. The values of he oal jagged areas end o converge and are close o he value of he oal jagged area obained using he minimum area algorihm. Thus, he muli-pso algorihm has excellen global search and local convergence for nodes a diverse posiions, improving he early convergence limiaion of he sandard PSO algorihm. Afer he jagged curve is calculaed using he muli-pso algorihm, he posiions of he nodes on he jagged curve are adjused. This adjusmen resuls in he jagged curve exhibiing a relaively smooh and curved ouline, achieving he required upsream-downsream proecion coordinaion. Paricle Swarm Paricle 1 Paricle 2 Paricle 3 Paricle 1 Paricle 2 Paricle 3 Paricle 1 Paricle 2 Paricle 3 Paricle 1 Paricle 2 Paricle 3 Iniial of 1, 1, 1, Table V Comparison of he oal jagged area. Iniial of Node c 1, 1, 1, 1, 1, 1, Toal Area of Iniial Toal Area Wih Sandard PSO Toal Area Wih Muli-PSO Toal Area Wih Minimum Area Mehod VI. CONCLUSION During he process of isolaing feeder fauls, he insufficien hardware compuing power of s necessiaes using he TCC mehod o calculae faul curves, which resuls in jagged curves and considerably high deecion errors. In his sudy, a muli-pso algorihm was used o improve he excessive flagging and failure o flag caused by he rule-of-humb mehods currenly used by Taipower. Firs, mahemaical opimizaion is conduced o esablish a mahemaical model for opimizing a se poins on he TCC. The objecive funcion is he minimized curve area, and he consrains are ha he se poins mus be wihin he inerval beween he fuse complee meling curve and he faul curve and ha he ime inerval beween he faul curve faul curve mus be a leas 20% of he pu ime value. The simulaion resuls indicae ha he muli-pso algorihm can improve he early convergence problems encounered in he sandard PSO algorihm. Nex, he jagged curves formed by nodes were simulaed a various posiions. Afer he muli-pso algorihm is execued o calculae he jagged curve and he node posiions on he jagged curve are adjused, he resuling jagged area is close o he moved nodes. This resuls in he jagged curve exhibiing relaively smooh and curved oulines, achieving he required upsream-downsream proecion coordinaion. This subsanially reduces malfuncions or erroneous nonoperaion in he breakers. Afer 2 years of esing, he proposed mehod is deermined o be able o effecively improve misjudgmen of fauls on he disribuion feeders by s. This increases he success rae of he exper sysem (FDIR). ACKNOWLEDGMENT The auhors graefully hank he assisance from he disribuion deparmen of Taipower o provide he valuable daa and he assisance of on-sie esing as well as excellen discussion. The financial suppor o his work by he Taipower is also highly appreciaed. REFERENCES [1] M. Meiqin, J. Meihong, D. Wei and L. Chang, Muli-objecive Economic Dispach Model for A Microgrid Considering Reliabiliy, IEEE Power Elecronics for Disribued Generaion Sysems (PEDG) on Inernaional Symposium, 20, pp [2] Qi Kang, MengChu Zhou, Jing An, Qidi Wu, Swarm Inelligence Approaches o Opimal Power Flow Problem Wih Disribued Generaor Failures in Power Neworks, IEEE Trans. Power on Auomaion Science and Engineering, vol., no. 2, pp , April [3] Mo N., Zou Z.Y., Chan K.W., Pong T.Y.G., Transien Sabiliy Consrained Opimal Power Flow Using Paricle Swarm Opimisaion, IET Gener. Disrib., pp , January [4] Das T.K., Venayagamoorhy G.K., Aliyu U.O., Bio-Inspired Algorihms for he Design of Muliple Opimal Power Sysem Sabilizers: SPPSO and BFA, IEEE Trans. Ind. Appl., vol. 44, no. 5, pp , Ocober [5] Communicaions Sysems Division, Taiwan Power Company. Tes Websie for Calculaing Opimal Seings. Taipower Operaions, Taipei, [6] Taiwan Power Company. Disribuion Technical Manual (25), Feeder Auomaion Engineering. Taipower Operaions, Taipei, [7] Taiwan Power Company, IED Specificaions. Taipower Operaions, Taipei, [8] Daneshyari M., Yen G.G., Consrained Muliple-Swarm Paricle Swarm Opimizaion Wihin a Culural Framework, IEEE Trans. on Sysem, Man, and Cyberneics- Par A: Sysem and Humans, vol. 42, no. 2, pp , March [9] Yen G.G., Wen Fung Leong, Dynamic Muliple Swarms in Muliobjecive Paricle Swarm Opimizaion, IEEE Trans. on Sysem, Man, and Cyberneics- Par A: Sysem and Humans, vol. 39, no. 4, pp , July [] Jurgen Schlabbach, Shor-Circui Currens, The Insiuion of Elecrical Engineers, London, Unied Kingdom, pp.1-96, [11] Hadi Saada, Power Sysem Analysis, McGraw-Hill inernaional Ediions, pp , [12] Dr.Frank Mercede, Faul Calculaions of Indusrial Commercial Power Sysems, The Insiue of Elecrical and Elecronics Engineers, Inc., ch.7, 12, 1994.

9 Inernaional Journal of Elecrical & Compuer Sciences IJECS-IJENS Vol:15 No:03 15 Ming-Yuan Cho(M 92) is wih he Deparmen of Elecrical Engineering, Naional Kaohsiung Universiy of Applied Sciences (KUAS), where he is currenly a disinguished professor associaed wih he Dean of College of Elecrical Engineering and Compuer Science a KUAS. He has also served as he chairman of Deparmen of Elecrical Engineering a KUAS from Augus, 2000 o July, His research ineress are smar grid, energy saving echnologies and arificial inelligen algorihm for power sysem applicaions. Shen-Wen Hsiao is currenly pursuing he Ph.D. degree a he insiue of elecrical engineering in Naional Kaohsiung Universiy of Applied Sciences, Taiwan. The pas couple of years have been an exceedingly busy and challenging ime for his sudy. His research ineress are opimizaion, suppor vecor machine and neural nework in ransmission and disribuion sysem for power sysem applicaions. Chia-Yuan Hsu was born in Kaohsiung, Taiwan, He received he M.S. degree in elecrical engineering from Kaohsiung Universiy of Applied Sciences, Kaohsiung, Taiwan, in He is now wih he Hsin-Ing disribuion disric, Taipower since His research ineress are power sysem conrol, SCADA and AI applicaions.