Journal of Theoretical and Applied Information Technology.

Transcription

1 DETECTION, CLASSIFICATION LOCALISATION, AND CONTROLLING OF VOLTAGE SWELLS USING IUPQC THROUGH WAVELET BASED NEURAL NETWORKS Dr. M.SUSHAMA, Dr G.TULASI RAM DAS, Dr. A. JAYA LAXMI Abstract Vast spread of sesitive loads i power systems results i icreasig susceptibility to power quality problems, which makes fast detectio ad classificatio ad localizatio algorithms a ecessity. A ew approach for power quality evet detectio usig Wavelet Multi Resolutio aalysis (MRA)is preseted i this paper. For Classificatio, Wavelet trasform is utilized to extract feature vectors for various PQ disturbaces based o the Multi Resolutio Aalysis (MRA). These feature vectors the are applied to the Neural Network system. For the compesatio of the Voltage Swell a Iterlie Uified Power Quality Coditioer(IUPQC) was employed. The complete was carried out usig MATLAB/ Simulik. Key words : Power Quality (PQ),Wavelet Trasforms(WT), Multi Resolutio Aalysis(MRA), Iterlie Power Quality Coditioer(IPQC)..INTRODUCTION The recet proliferatio of electroic equipmet ad microprocessor-based cotrols has caused electric utilities to redefie PQ i terms of the quality of voltage supply rather tha availability of power. Recommeded Practice for Moitorig Electric Power Quality, has defied a set of termiologies ad their characteristics to describe the electrical eviromet i terms of voltage quality. Voltage sag is a short-duratio decrease of the Root Mea Square (RMS) voltage, lastig from.5 cycle to two miutes i duratio (fig.). These evets are caused by faults o the power system or by the startig curret of a relatively large motor or other large load. Typically for trasmissio faults, these voltage disturbaces last for fractios of a secod ( / secod), which represets the total fault-clearig time for trasmissio faults. However, these mometary evets ca cause a complete shutdow of plat wise processes, which may take hours to retur to ormal operatio(as stated by Duga 24). A voltage swell may accompay voltage sag. A voltage swell occurs whe a sigle lie-togroud fault o the system results i a temporary voltage rise o the ufaulted phases (fig.). Removig a large load or addig a large capacitor bak ca also cause voltage swells, but these evets ted to cause loger duratio chages i the voltage magitude ad will usually be classified as log-duratio variatios (fig). Figure Example waveforms for short-duratio voltage variatios Electric utilities must assess the preset value of the power before takig ay quality improvemet actios. Therefore, detectio of power quality disturbaces has become a sigificat issue. Voltage variatio caused by fault coditios ad the eergizatio of large loads, where high startig currets are ivolved. The faults ca cause a drop, rise ad supply void i the supply voltage, are also kow as sag, swell ad iterruptios respectively. 2. WAVELET TRANSFORM A sigal ca be represeted i the frequecy domai by its Fourier trasform which is writte as : N- F[k] = ƒ [] e -j(2 πk/n) N =, k=,, N-... () where ƒ[] is the discrete-time sigal. F[k] is its frequecy domai represetatio. The wavelet trasform has such a zoomig property. I cotrast to the Fourier trasform, the wavelet trasform does ot look for circular 8

2 frequecies but rather for detail sizes a at a certai time t. Istead of detail sizes, we will also speak of scale factors, both otios will be used equivaletly[]. High frequecies correspod to small details ad vice versa, thus, whe comparig wavelet with Fourier trasforms we have to take ito accout that frequecies ad detail sizes are iversely proportioal to each other: There exists a costat β such that (2) β a = ω Joural of Theoretical ad Applied Iformatio Techology Wavelet tool provides a way of aalyzig a sigal both i time ad frequecy domais. If we deote ƒ as a fuctio defied o the whole real lie, the, for a suitably chose mother wavelet fuctio Ψ,ƒ ca be expaded as j / 2 j f ( t) = c ϕ( t k) + d 2 ψ (2 t k) k k = k = j= j, k (3) where Ψ (2 j/2 ) & Ψ(2 j t-k) are all orthogoal to each other. The coefficiets w jk gives iformatio about the behavior of the fuctio ƒ cocetratig o the effects of scale aroud 2 -j ear time t x 2 -j. j determies the scale or the frequecy rage of each wavelet basis fuctio ψ (t) ad k determies the time traslatio. c k are called approximatio coefficiets ad, d j,k are called wavelet coefficiets[]. I the above expasio, the first summatio gives a fuctio that is a low-resolutio or coarse approximatio of f(t). For each idex j i the secod summatio, a higher or fier resolutio fuctio is added, which adds ew details. Geerally, the wavelet trasform ca be represeted by filter-bak operatios where a series of liear low- ad high-pass filters decompose the sigals ito successive details ad approximatios compoets. The filter operatios are followed by dyadic dow samplig the itermediate output sigals, that is, at each level of decompositio every other sample is discarded. This wavelet decompositio of a fuctio is closely related to a similar decompositio (the discrete wavelet trasform, DWT) of sigal observed i discrete time[4]. Fig.2 Implemetatio of DWT usig filter baks. Figure 2 shows the tree structure implemetatio of filter baks for oedimesioal DWT, where g() stads for the high-pass filters ad h() for the low-pass filters ad the arrows stad for the dow samplig process. There are some importat properties of these filters 2 2 h( ) = ad g ( ) = (4) h ( ) = 2 ad g ( ) = (5) Filter g() is a alteratig flip of the filter h(), which meas there is a odd iteger N such that, g ( ) = ( ) h ( N ).(6) Daubechies gives a detailed discussio about the characteristics of these filters ad how to costruct them. The decompositio procedure starts with passig a sigal through these filters. The approximatios are the low-frequecy compoets of the time series ad the details are the high-frequecy compoets. Multi resolutio aalysis leads to a hierarchical ad fast scheme. This ca be implemeted by a set of successive filter baks as show i Fig.2, where h() ad g() are the low-pass ad high-pass filters as defied i (4) (6). 2 meas the dow samplig with a factor of 2,k is the coefficiet idex at each decompositio level. 9

3 3. MULTI RESOLUTION ANALYSIS (MRA) Cosiderig the filter bak implemetatio i Fig.3, the relatioship of the approximatio coefficiets ad detail coefficiets betwee two adjacet levels are give as usig the followig equatios[2]: ca j (k) = h 2k ) ca j ( ) (.(7) cd j (k) = g ( 2k ) ca j ( ).(8) where ca j ad cd j represet the approximatio coefficiets ad detail coefficiets of the sigal at level j, respectively. I this way, the decompositio coefficiets of MRA aalysis ca be expressed as, [A ] [ca, cd ] [ca 2, cd 2, cd ] [ca 3, cd 3, cd 2, cd ] which correspod to the decompositio of sigal x(t) as, x(t) = A (t) + D (t) = A 2 (t) + D 2 (t) + D (t) = A 3 (t) + D 3 (t) + D 2 (t) + D (t) = where A i (t) is called the approximatio at level i, ad D i (t) is called the detail at level i. Time frequecy aalysis was applied to detect the presece of the trasiets added to voltage sigals by ay disturbace. These trasiets ca be extracted from the wavelet detail compoets. The ability of the wavelet trasform to detect particular trasiets i the voltage sigals depeds o the type of the wavelet family, its order ad the level of decompositio, that is, the umber of detail compoets extracted. I wavelet applicatios, differet basis fuctios were proposed ad selected. No sigle wavelet trasform has a statistically sigificat advatage over other wavelets i performace of PQ classificatio. I the proposed scheme, the Db4 wavelets were selected as the wavelet basis fuctio for the detectio ad classificatio of voltage disturbace. The wavelet aalysis block trasforms the distorted sigal ito differet time-frequecy scales detectig the disturbaces preset i the power sigal. The wavelet trasform (WT) uses the wavelet fuctio Ψ ad scalig fuctio Φ to perform simultaeously the MRA decompositio ad recostructio of the measured sigal. The wavelet fuctio Ψ will geerate the high frequecy compoets (details) ad scalig fuctio Φ will geerate the low frequecy compoets (approximatios) of the distorted sigal. Discrete Wavelet Trasform (DWT) is the basic tool for feature extractio[2]. DWT is the discrete couterpart of the Cotiuous Wavelet Trasform (CWT). The CWT of a cotiuous time sigal x (t) is defied as α CWTψ x( a, b) =.(9) α x( t) * ψ a, b( t) dt, a, b R, a, t b a *, b t = a a * where, ψ ( ) ψ..() The fuctio Ψ (t) is the mother wavelet, ad the asterisk deotes a complex cojugate. a ad b are the scalig ad traslatig parameters respectively. The sampled sigal x k is used to replace the CWT of x t such that * DWTψ x( m, ) = k xkψ m, ( k),.() where, m * * k ba ψ m, ( k) = ψ. m m a a.(2) Both the scalig factor b a m a ad the shiftig factor are fuctios of the iteger parameter m, where m ad are scalig ad samplig umbers respectively ad m =,, 2, By selectig a = 2 ad b =, a represetatio of ay sigal x k at various resolutio levels ca be developed by usig the MRA. 4 FEATURE EXTRACTION Stadard deviatio multi resolutio aalysis curve (Std MRA) has the ability to quatify the magitude of variatio withi the sigal. The extracted features help to distiguish oe disturbace evet from aother. MRA ca detect m 2

4 ad diagose defects ad provide early warig of power quality problems. I power quality problem, the graph of stadard deviatio of multi resolutio aalysis is very similar i some cases such as swell, otchig, iterruptio ad voltage swell as show i fig. 3 (a) (b) Fig.3(a),(b)Stadard deviatio MRA (Std _MRA) Of Voltage Swell 4.. WAVELET BASED FEATURE EXTRACTION I order to reduce the feature dimesio, we will ot directly use the detail D i (t) ad approximate A i (t) iformatio for future traiig ad testig[3][7]. Istead, we propose to use the eergy at each decompositio level as a ew iput variable for Back propagatio classificatio. The eergy at each decompositio level is calculated usig the followig equatios: N 2 ED. = D, i = l.(3) x j= N j = ij 2 EA l = A ij.(4) where i=,2, l is the wavelet decompositio level from level to level l. N is the umber of the coefficiets of detail or approximate at each decompositio level. ED i is the eergy of the detail at decompositio level i ad EA l is the eergy of the approximate at decompositio level l. I this way, for a l level wavelet decompositio, we costruct a (l+) dimesioal feature vector for future aalysis. Fig.4 shows data flow i the proposed wavelet feature extractio. The power sigal with disturbace whe subjected to DWT will geerate a discotiuous state at the start ad ed poits of the disturbace duratio. For each of the disturbace, the DWT coefficiets geerated have variatios which are used to recogize the various power sigal disturbace ad thereby classifyig the differet power quality problems(fig.4)[7]. By applyig DWT, the distorted sigal ca be mapped ito the wavelet domai ad represeted by a set of wavelet coefficiets. There are differet wavelets that ca be used to decompose the distorted sigal ad extract the feature vector. Here the Daubechies Db4 wavelet fuctio is used to decompose the sigal by DWT. Sice maximum eergy localizatio is obtaied usig db4 ad db8 whe compared to the other type of wavelets db4 is used. The parameters of the voltage waveforms durig Power Quality (PQ) evets are statistically differet from those that are calculated durig a evet free time period. This statistical differece is used for effective detectio of the PQ evets like sag/swell, mometary iterruptio, harmoics etc The statistical behavior of the feature vectors is obtaied. Local statistical properties of the waveform vary with the PQ evets. PQ disturbaces Pure Siusoidal Mometary Swell Temporary Swell Log-term Swell (Over voltage) Class Symbol C C2 C3 C4 Table. Model x(t) = si(ωt) x(t) = A(+α(u(tt )-u(tt 2 )))si(ωt) x(t) = A(+α(u(tt )-u(tt 2 )))si(ωt) x(t) = A(+α(u(tt )-u(tt 2 )))si(ωt) Parameters. α.4. α.4. α.2 2

5 VOLTAGE SWELLS 6.. PURE SINUSOIDAL It is the ideal voltage waveform geerated by pure siusoidal sigal. The sigal is geerated (fig 5) at 5 Hz havig p.u magitude as show i fig.(5). Fig. 4. Wavelet based feature extractio. 5. VOLTAGE SWELL CONDITIONS OF POWER QUALITY EVENTS: Several typical PQ disturbaces, especially voltage swells are take ito cosideratio i this paper. Usig MATLAB 7.., the most commoly occurrig disturbaces are iitially simulated. The categories that are simulated are ormal siusoid, voltage swell, further categorized as Mometary, Temporary ad log term swell. 6. GENERATION OF VOLTAGE SWELL SIGNALS These sigals geerated are sampled at a frequecy of 4 khz. The uique attributes for each disturbace type are used(table.) ad allowed to chage radomly, withi specified limits, i order to create differet disturbaces. Fig.5 Pure Siusoidal Wave form 6.2. VOLTAGE SWELL Voltage swell is described as a drop of -9% of the rated system voltage lastig for half a cycle to oe miute. The causes of voltage swell are :. Voltage swell are caused by system faults 2. It ca also be caused by eergisatio of heavy loads. This sigal geerated i fig.6 (a), (b) are from the model equatio x(t) = A(+α(u(t-t )-u(t-t 2 )))si(ωt) where, t <t 2,, u(t) =, t. α.9;t t 2 -t 9T =, t <. ad the decompositio up to four levels (a 4, d 4 ) usig Db4. Voltage(V) iput sigal -SWELL with 3%load Time(msec) 22

6 (a) Fig.7.a.5 iput sigal -SWELL with 2%load.5 Voltage(V) Time(msec) Fig.7.b (b) Fig 6. (a) Voltage Swell Sigal (b) Decomposed Sigal usig Wavelet (Db4) Voltage swell is agai classified as Mometaeous, Temporary ad log term over voltages depedig upo the time duratio of the evet[4]. The stadards of limitatios as per IEEE are give i table 2 as follows: Table 2. Categories of power quality variatio (Voltage Swells) Istitute of Electrical ad Electroics Egieers (IEEE) Type of Swell Time duratio Typical amplitude Voltage(V) iput sigal -SWELL with 3%load Time(msec) Fig.7.c Fig 7 (a),(b),(c) Voltage Swells Table. VOLTAGE SWELLS (o ext page) Mometaeous swell Temporary swell Log-term Over Voltage 3 cycles to 3 sec. p.u to.4 p.u 3 sec to mi. p.u to.4 p.u > mi. p.u to.2 p.u The %, 2%, 3% swell disturbace lastig for 5cycles are show i Fig 7(a),(b),(c)..5.5 iput sigal -SWELL with %load Fig 8. Feature Extractio & Target vector for the Neural Network. 7.Detectio of Voltage Swell : Voltage(V) Time(msec) 23

7 Geerally speakig, feed-forward etworks are static, that is, they produce oly oe set of output values rather tha a sequece of values from a give iput. Feed- forward etworks are memory-less i the sese that their respose to a iput is idepedet of the previous etwork state. Recurret or feedback etworks, o the other had, are dyamic systems. Whe a ew iput patter is preseted, the euro outputs are computed. Because of the feedback paths, the iputs to each euro are the modified, which leads the etwork to eter a ew state. Differet etwork architectures require appropriate learig algorithms. Fig 9 Detectio of Istataeous Swell I fig 9. the detectio of Istataeous type swell was show, where the fall i voltage is 35.5% ad the duratio of the fault id 2 msec. CLA-- SS C2 C3 C4 C C 8.. CLASSIFICATION USING BACK PROPAGATION: A sigle-layer etwork of S logsig euros havig R iputs is show below i figure. Feed forward etworks ofte have oe or more hidde layers of sigmoid euros followed by a output layer of liear euros[]. Multiple layers of euros with oliear trasfer fuctios allow the etwork to lear oliear ad liear relatioships betwee iput ad output vectors. The liear output layer lets the etwork produce values outside the rage to +. O the other had, if you wat to costrai the outputs of a etwork (such as betwee ad ), the the output layer should use a sigmoid trasfer fuctio (such as logsig). C2 8 7 C C ARTIFICIAL NEURAL NETWORK: ANN s ca be viewed as weighted directed graphs i which artificial euros are odes ad directed edges are coectios betwee euro outputs ad euro iputs. Based o the coectio patter (architecture), ANN s ca be grouped ito two categories: * Feed-forward etworks, i which graphs have o loops * Recurret (or feedback) etworks, i which loops occur because of feedback coectios. Fig. Sigle layer etwork 9. SIMULATION AND ANALYSIS: The simulatio data was geerated i MATLAB based o the model i table. All the four classes (C C4) of differet PQ (Voltage Sag type) disturbaces, amed udisturbed siusoid (ormal), sag ad its differet categories. Table I gives the sigal geeratio models ad their 24

8 cotrol parameters. Sevety five cases of each class with differet parameters were geerated for traiig ad aother 25 cases were geerated for testig. Both the traiig ad testig sigals are sampled at 2 poits/cycle ad the ormal frequecy is 5 Hz. Fiftee power frequecy cycles which cotai the disturbace are used for a total of 3 poits. Daubechies4 (Db4) wavelets with four levels of decompositio were used for aalysis (l=4). Based o the feature extractio show above, 4-dimesioal feature sets for traiig ad testig data were costructed. The dimesios here describe differet features resultig from the wavelet trasform, that is to say, the total size of the traiig data or testig data set is x4, where 4 comes from cases per class multiplied by 4 classes ad 4 is the dimesio of the feature size of each case. All data sets were scaled to the rage of ( 2) before beig applied to Feedforward back propagatio etwork for traiig ad testig[3]. The results are tabulated for all the 4 evets i table 4. C Pure Siusoidal C3 Temporary Swell C2 Mometaeous Swell C4 Log-term Swell (over voltage). Table 3 Simulatio results certai travellig wave from the locator positio or with aalyzig the geerated trasiets due to the fault occurrece. Impedace measuremet schemes are classified whether they deped o the data from oe or both lie eds. Each category ca be the classified accordig to the cosidered lie model durig the derivatio method usig either simpler (lumped) models or detailed (distributed parameters) oes. Accordig to travellig wave theory, voltage ad curret travellig waves appear o the lie whe fault occurs. The fault geerated travellig waves cotai sufficiet fault iformatio that ca be used for high-speed fault idetificatio ad lie protectio[4]. I AC trasmissio lies, the amplitude of fault geerated travellig waves chages with the voltage agles. The pure frequecy domai based methods are ot suitable for the timevaryig trasiets ad the pure time domai based methods are very easily iflueced by oise. The wavelet trasform provides a ew approach for aalyzig time-varyig trasiets. It has the capability of aalyzig sigals simultaeously i time ad frequecy domai. Moreover, it ca adjust aalysis widows automatically accordig to frequecy, amely, shorter widows for higher frequecy ad vice visa. Hece it is suitable for characteristic idetificatio ad travellig wave protectio. The results show that the wavelet techiques lead to a ew way for the fault idetificatio ad the protectio... Wavelet Modulus Maxima (WMM): It ca be see that durig fault period, the WMM udergoes sudde chages ad it rises to a large value durig the fault period ad is of low value i absece of fault. By observig the amplitude of WMM, ac fault ca be kow[5]. I ormal operatio, WMM is of low value ad there are o sudde chages which idicate there is o fault. Fig. Traiig the BPNN. LOCALIZATION OF THE EVENTS USING WAVELET MODULUS MAXIMA (WMM): Geerally speakig, fault locatio methods ca be classified ito two basic groups, travelig wave-based schemes ad impedace measuremet-based oes. Travellig wave schemes ca be used either with ijectig a Fig.2. Applicatio of WMM to Travellig wave of Voltage sag 25

9 . CONTROLLING THE VOLTAGE SWELL USING INTERLINE UNIFIED POWER QUALITY CONDITIONER(IPQC) : A UPQC cosists of a series voltage-source coverter (VSC) ad a shut VSC both joied together by a commo dc bus. It is demostrated how this device is coected betwee two idepedet feeders to regulate the bus voltage of oe of the feeders while regulatig the voltage across a sesitive load i the other feeder(fig3). Sice the UPQC is coected betwee two differet feeders (lies), this coectio of the UPQC will be called a iterlie UPQC (IPQC)[6]. The structure, cotrol ad capability of the IPQC are discussed i this paper. the sesitive load is fully protected agaist sag/swell ad iterruptio. The sesitive load is usually a part of a process idustry where iterruptios result i severe ecoomic loss. Therefore, the cost of the series part of IPQC must be balaced agaist cost of iterruptios based o past reliability idices (e.g., CAIFI,CAIDI). It is expected that a part of IPQC cost ca be recovered i 5 years by chargig higher tariff for the protected lie. Furthermore, the regulated bus B- ca supply several customers who are also protected agaist sag ad swell. The remaiig part of the IPQC cost ca be recovered by chargig higher tariff to this class of customers. Such detailed aalysis is required for each IPQC istallatio. Fig 5 Voltage swell applied at feeder without IPQC Fig 3 Sigle lie diagram of a IPQC Fig. 4 Typical IUPQC coected i a distributed System i a distributio System The performace of the IPQC has bee evaluated uder various disturbace coditios such as voltage swell i either feeder, fault i oe of the feeders ad load chage. It has bee show i fig.4 that i case of a voltage swell, the phase agle of the bus voltage i which the series coected VSC plays a importat role as it gives the measure of the reactive power required by the load. The IPQC ca mitigate a voltage swell of about.35 p.u. (9 kv to 2.5 kv) i Feeder-(fig 6) ad.6 p.u. (i.e.,9 kv to kv) i Feeder-2 for log duratio(fig. 6). I the IPQC cofiguratio discussed i this paper, Fig 6 Compesated Voltage Sigals at feeder &2 with IPQC 26

10 2. CONCLUSIONS: I this paper oe type of voltage sigal disturbace, called voltage swell was detected usig wavelet decompositio techique. For differet types of voltage swell coditios like mometary swell, temporary swell ad log term over voltage swell, the classificatio was doe with the help of feature extractio usig Multi Resolutio Aalysis (MRA) ad Feed Forward Back Propagatio Neural Network Traiig Algorithm. The most importat part of the work is to locate the fault beig accurately doe usig Travellig Wave method ad Wavelet Modulus Maxima (WMM). After the process of detectio, classificatio ad localizatio, it is essetial to cotrol or compesate the voltage sigal disturbace usig a proper cotrollig device called Uified Power Quality Coditioer (UPQC). O the distributio side two feeders F & F2 were cosidered. I feeder, Voltage Swell has bee geerated. The cotrollig has bee observed i both the feeders usig Iterlie Uified Power Quality Coditioer (IPQC). All the work was carried out usig MATLAB/Simulik (Wavelet ad Neural Network Toolboxes). The lies are protected satisfactorily from swell disturbaces by usig IPQC. REFERENCES: []. W. R. A. Ibrahim ad M. M. Morcos, Artificial itelligece ad advaced mathematical tools for power quality applicatios: A survey, IEEE Tras. Power Del., vol. 7, o. 2, pp , Apr. 22. [2]. S. Satoso, J. Lamoree, W. M. Grady, E. J. Powers, ad S. C. Bhatt, A scalable PQ evet idetificatio system, IEEE Tras. power Del., vol.5, o. 4, pp , Apr. 2. [3]. S. Satoso, E. J. Powers,W. M. Grady, ad A. C. Parsos, Power quality disturbace waveform recogitio usig wavelet-based eural classifier part I: Theoretical foudatio, IEEE Tras. Power Del., vol. 5, o., pp , Ja. 2. [4]. S. Satoso, E. J. Powers,W. M. Grady, ad A. C. Parsos, Power quality disturbace waveform recogitio usig wavelet-based eural classifier- part II: Applicatio, IEEE Tras. Power Del., vol. 5, o., pp , Ja. 2. [5]. C. S. Burrus, R. A. Gopiath, ad H. Guo, Itroductio to Wavelets ad Wavelet Trasforms: A Primer.Eglewood Cliffs, NJ: Pretice-Hall, 998. [6]. S. Mallat, A theory for multiresolutio sigal decompositio: The wavelet represetatio, IEEE tras Patter Aal. Mach. Itell., vol, o. 7, pp , Jul [7]. Self-Orgaizig Learig Array System for Power Quality Classificatio Based o Wavelet Trasform by HaiboHe, Studet Member, IEEE, ad Jausz A. Starzyk,Seior Member, IEEE. [8]. Power Sigal Disturbace Classificatio Usig Wavelet Based Neural Network by S.Suja, Jovitha Jerome2-SERBIAN JOURNAL OF ELECTRICAL ENGINEERING Vol. 4, No., Jue 27, [9]. W. R. A. Ibrahim ad M. M. Morcos, Artificial itelligece ad advaced mathematical tools for power quality applicatios: A survey, IEEE Tras. Power Del., vol. 7, o. 2, pp , Apr. 22. []. R. Aggarwal ad C. H. Kim, "Wavelet Trasforms i Power Systems:Part Geeral Itroductio to the Wavelet rasforms," Power Egieerig Joural, vol.4, pp. 8-87, Apr. 2. []. S. Mallat, A Wavelet Tour Of Sigal Processig, 2d Editio, Califoria, USA:Academic Press, 999. [2]. I. Daubechies, "The wavelet trasform, time-frequecy localizatio ad sigal aalysis," IEEE Trasactio o Iformatio Theory, vol. 36, pp. 96-5, Sep.99.. [3]. A. S. Bretas, L. O. Pires, M. Moreto ad R. H. Salim, "A BP Neural Network Based Techique for HIF Detectio ad Locatio o Distributio Systems with Distributed Geeratio," Lecture Notes i Computer Sciece, vol. 44, pp , Sep. 26. [4]. Mallat, S. ad Hwag, W. L., 992, Sigularity detectio ad processig, IEEE Tras. o Iformatio Theory, 38-2, [5]. Dog, X. Zh., Ge, Y. Zh. ad Xu, B. y., 2, Fault positio relay based o curret travellig waves [6]. ad wavelets, IEEE PES 2 Witer Meetig, [7]. A. Ghosh, A. K. Jidal, ad A. Joshi, A uified power quality coditioer for voltage regulatio of critical load bus, i Proc. IEEE Power Eg. Soc. Geeral Meetig, Dever, CO, Ju. 6, 24 27

11 Biographies of Authors Dr. G.Tulasi Ram Das bor i the year 96, i Hyderabad Adhra Pradesh, Idia ad received his B.Tech degree is i Electrical & Electroics Egieerig from J.N.T.U College of Egieerig, Hyderabad, Idia i 983 ad M.E with Idustrial Drives & Cotrol from O.U College of Egieerig, Hyderabad, Idia i 986. He received the PhD, degree from the Idia Istitute of Techology, Madras, Idia i 996. He is havig 22 years of teachig ad research experiece. He is curretly Professor at J.N.T.U College of Egieerig, Hyderabad- 5 72, Idia. He held positios of Head, Departmet of Electrical ad Electroics Egg., Vice-Pricipal ad Pricipal of JNTU College of Egieerig, Hyderabad. His Research iterests are Power Electroics, Idustrial Drives & FACTS Devices. He has supervised 7 Ph.D. theses ad 7 Ph.D. theses are uder progress. He has published/preseted 7 techical research papers i atioal ad iteratioal cofereces ad jourals. He has visited coutries amely, Sigapore, Malaysia ad Easter States of US. Dr. A. Jaya laxmi was bor i Mahaboob Nagar District, Adhra Pradesh, o She completed her B.Tech. (EEE) from Osmaia Uiversity College of Egieerig, Hyderabad i 99, M. Tech.(Power Systems) from REC Waragal, Adhra Pradesh i 996 ad completed Ph.D.(Power Quality) from Jawaharlal Nehru Techological Uiversity College of Egieerig, Hyderabad i 27. She has five years of Idustrial experiece ad 8 years of teachig experiece. She has worked as Visitig Faculty at Osmaia Uiversity College of Egieerig, Hyderabad ad is presetly workig as Associate Professor, JNTU College of Egieerig, Hyderabad. She has oe Iteratioal Joural to her credit. She has 8 Iteratioal ad 5 Natioal papers published i various cofereces held at Idia ad also abroad. Her research iterests are Neural Networks, Power Systems & Power Quality. She was awarded Best Techical Paper Award for Electrical Egieerig i Istitutio of Electrical Egieers i the year 26. Dr. A. Jaya laxmi is a Member of Istitutio of Electrical Egieers Calcutta (M.I.E) ad also Member of Idia Society of Techical Educatio (M.I.S.T.E). Dr. M.Sushama, bor o 8 th Feb 973, i Nalgoda, a small tow ear Nagarjua Sagar, A.P., Idia.Obtaied her B.Tech degree i 993 ad M.Tech degree i 23 with a specializatio i Electrical Power Systems from JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY, INDIA. Joied as Lecturer i the Departmet of EEE, JNTU College of Egg., Aatapur, i the year 995. I the year 996, redesigated as Assistat Professor. Presetly workig as Associate Professor, Electrical & Electroics Egieerig i the Departmet of EEE, JNTU College of Egieerig, Kukatpally, Hyderabad. She had 3 years of teachig experiece. Durig her teachig career she taught various subjects like C laguage & Data Structures, Microprocessors ad Micro cotrollers etc., Doig research i Power Quality usig Wavelets, she published 5 iteratioal papers i various IEEE sposored cofereces. 28