ApplicationofWaveletstoAnalyzePowerSystemTransients Ö.N.Gerek*,D.G.Ece, DepartmentofElectrical-ElectronicsEngineering,AnadoluUniversity,Eskişehir,Turkey Abstract Powersystemtransientsremaintobeanissueforenergydeliveryefficiency.Waveletshavebeenauseful tooltoanalyzevarioussignalsandwaveformsforitsgoodtime-frequencyproperties,includingtheanalysis ofpowersystemtransientsandpowerqualityproblems.Inmostoftheresearches,popularwaveletsare selected, and it has been observed that the wavelet decomposition of transient waveforms at various resolutionsprovideafairemphasisonthedetectionandclassificationofpowersystemabnormalities.Since thetransformcoefficientsdepictthetransientwaveformcharacteristicsinatime-windowedmanner,these coefficientsindicatethetimeperiodoverwhichatransienteventoccurs.Nevertheless,thechoiceordesign ofthewaveletforthisparticularpurposehasnotbeenaddressedbefore.Thispaperprovidesresultsfor variouschoicesofwavelets,andproposesaspecificfilterforthesubbanddecompositionanddetectionin subbanddecompositionofvoltagewaveformsunderstagedincipientfaultsandspeeddrivestarting. ©2016IEESE.Allrightsreserved. 1.Introduction TheImprovingcomputeranddigitalprocessingtechnologieshaveenabledmorepracticaluse of automated computer based systems which digitally analyze the power system transients by filtering and frequency domain methods. Recently wavelet analysis has been a very popular methodinthisaspect[1]-[12].Itsabilitytodepictfrequencycontentatdifferentscalestogether withpreservingtimelocalityinducesagreatpotentialforsignalanalysis.Unliketheshorttime Fourier Transform (STFT), the wavelets can be selected, generated, or designed for satisfying specificpropertiesoftheclassofsignalsunderconsideration. Multiplelevelwaveletdecompositionisperformedtoselecttheenergycontentofthesignalina specific frequency band. In this aspect, wavelets can be thought of as practical tools for determiningthespectrumofthewaveforminatime-windowedmanner.Infact,mostresearchers intheareaofpowerqualityanalysishaveconsideredandusedthewellknownwaveletsasaway ofsplittingthefrequencyspectrumofthewaveformusinghigh-andlow-passfilterpairs.Ifonly this property of the wavelet filter banks are considered, the choice of the wavelet becomes irrelevant.Whentheideaistoselectgoodlow-andhigh-passfilters,thereislittlereasontoselect the wavelets as Daubechies or Morlets, etc. However, wavelets have many other time-domain properties. Inthiswork,weinvestigatedwaysofgeneratingwaveletfilterbanksdependingontheirspectral andtimedomainproperties,andcomparedthemtostandardwavelets.Twomethodsusedhere to generate filter banks are; selection of a half band filter-pair due to its good band splitting property,andselectingafilterwhichapproximatesthewaveformwithoutatransient.Filterswere usedtoanalyzereallifecurrentwaveformdataacquiredfromaphaseofanexperimentallowvoltagesystemloadedwithanadjustablespeeddrive(ASD),resistiveandinductiveloadbanks. Phase-to-phasestagedincipientfaultcurrentandASDstartingcurrentwaveformsareanalyzed. It was observed that both filter banks generated in this work have on par, or better detection propertiesthanthewellknownwavelets. 2.Half-BandFilterPairsversusDaubechiesFilters Amajorobservationbetweennormalandfault-contaminatedvoltagesistheavailabilityversus unavailabilityofparticularharmonicsasseeninFigure1. * Correspondingauthor:ongerek@anadolu.edu.edu.tr Gerek&Ece/8thInternationalEgeEnergySymposiumandExhibition-2016 Fig1:(a)IncipientfaultcurrentwaveformofanASD,(b)Powerspectrumofthenoiselesspart, (c)Powerspectrumoffaultycurrent. Selectinglow-passandhigh-passfilterpairsisareasonableapproachtodistinguishbetween frequencybandcharacteristicsofthetwospectra.Indeed,inmostoftherecentworks,wavelets withgoodfrequencypropertieshavebeenused.However,onecanselectbetterfiltersthanthe readilyavailablefilters(i.e.,Daubechies,Morlet,etc.)Mostofthepopularwaveletsaredesigned forgeneralsignalprocessingpurposes,andtheymaylacksomeofthepropertieswhichmaybe desirable for analysis of power system transients. For example, when incipient fault occurs, frequencypeakscorrespondingto5thand6thharmonicsareemphasizedmoreinthenoisysignal. Forillustrativepurposes,wecomparedthe15-taphalfbandlow-pass(Eq.1)andhigh-passfilter pairwiththeDaubechies-4filterbank. [0.0074,0,0.0328,0,0.1076,0,0.4352,0.7083,0.4352,0,0.1076,0,0.0328,0,0.0074](1) Inananalysisof3-levelbalancedtreedecomposition,weobservethatthetwofilterbankcases provide exactly the same subband sample variation for the eight subbands. The observation is valid for both arcing faults and ASD motor start up. Hence, it is concluded that half band or Daubechies filters do not provide any improvement for event detection. The experiments are repeated for longer (hence sharper) half band filters and longer Daubechies wavelets, and comparativeresultsdidnotchange.Thisobservationalsoimpliesthatwaveletsgoodforspecific applicationsmaynotnecessarilycorrespondtogoodall-purposewavelets. 3.DesignofWaveletsforPowerTransientSignalAnalysis In this work, we have adopted the Daubechies-type wavelet generation strategy to come up withanalysisandsynthesisfilters.Sincetheapproximationsubspaceisgeneratedbythescaling function,wehavestartedwithscalingfunction(low-pass)filterdesignbymimickingaperiodof normalvoltagewavecycle,andincorporatedaspectralzeroat𝜔 = 𝜋,asrequiredforregularity. 2 Gerek&Ece/8thInternationalEgeEnergySymposiumandExhibition-2016 Consequently,thehighpassfilterwasgeneratedtocomplementtheapproximationwithaspectral zeroat𝜔 = 0.Theselectedscalingfunction(thefilter𝐻' )isshowninFigure2. Fig2:Thelow-passfilterthatmimicsanormalvoltagewaveform,itself. Again, wavelet decompositions are obtained and high frequency bands are investigated. In Fig.3(a),wecanseethattheapproximationwaveformcontainsmostofthesignalpower,whereas thedetailsignalsatvariousstagesarerathersmallforthenormalwaveform.Ontheotherhand, forthewaveformwithincipientfault(Fig.3(b)),thedetailsignalsbecomelargeranditcanbe better discriminated from the event free case than the detail signals using, for example, Daubechieswavelets(Fig.3(c)). Fig3:Pyramiddecompositionof(a)normalwaveform,(b)incipientfaultwaveformusingfilterinFig.2, and(c)usingDaubechies-6wavelet. Thesecondmethodtoobtainthepyramiddecompositionlowpassfilterisbasedontheadaptive predictionofthetransientwaveformsamplesfromitsadjacentsamplesintime.Forthispurpose, aLeastSquarestypeofadaptationwasappliedtothepartsofthewaveformwithoutatransient. Thecoefficientsoftheresultingnon-causal13-tapFIRfilterwasobtainedas [0.0029,0.0108,0.0297,0.0597,0.1066,0.3303,0.6836,0.3164,0.1201,0.0356,0.0332,0.0006,0.0049] Whenthisfilterisusedwithreallifesignalscontaininganincipientfault,weobtaintheresultsas showninFigure4(a).AswecanseeinFigure4(b),thisfilteremphasizesthefaultregionsbetter thantheDaubechies-6filterbankacrosssecond,third,andfourthscales. 3 Gerek&Ece/8thInternationalEgeEnergySymposiumandExhibition-2016 Fig4:Decompositionoftheincipientfaultwaveformusing(a)adaptedfilterbankand(b)Daubechies-6 wavelet. 4.Conclusions Inthiswork,weillustratedanattempttoobtainwaveletfiltersspecificallyoptimizedforpower system transient waveforms. Although it is true that wavelets constitute a nice way to detect abnormalbehaviorinsignals,ratherlittleattentionispaidontheselection,betteryet,thedesign of the wavelet suitable for the power system transient analysis. We proposed various ways to selectordesignsubbandfilterbanksforwaveletanalysis.Forthereallifevoltagedatasamples, weobtainedsuccessfulresultsthanconventionalandpopularwaveletsintermsofmoreapparent subbandenergyforPQfaultsituations.Theresultsshowthatmoreattentionneedstobepaidon thesignalcharacteristicsforselectingorgeneratingwaveletsforbetteroverallperformance. References [1] [2] [3] [4] [5] S. Santoso, W. M. Grady, E. J. Powers, J. Lamoree, and S. C. Bhatt, "Characterization of Distribution Power Quality Events with Fourier and Wavelet Transforms," IEEE Trans. PowerDelivery,vol.15,no.1,Jan.2000,pp.247-254. T.B.LittlerandD.J.Morrow,"WaveletsfortheAnalysisandCompressionofPowerSystem Disturbances,"IEEETrans.PowerDelivery,vol.14,no.2,April1999,pp.358-364. O. Poisson, P. Rioual, and M. Meunier, "New Signal Processing Tools Applied to Power QualityAnalysis",IEEETrans.PowerDelivery,vol.14,no.2,April1999,pp.561-566. S.J.Huang,C.T.Hsieh,andC.L.Huang,"ApplicationofMorletWaveletstoSupervisePower SytemDisturbances",IEEETrans.PowerDelivery,vol.14,no.1,Jan.1999,pp.235-243. S. J. Huang, and C. T. Hsieh, "High-Impedance Fault Detection Utilizing a Morlet Wavelet 4 Gerek&Ece/8thInternationalEgeEnergySymposiumandExhibition-2016 TransformApproach",IEEETrans.PowerDelivery,vol.14,no.4,Oct.1999,pp.1401-1410. [6] [7] [8] [9] T.Zheng,E.B.Makram,andA.A.Girgis,"PowerSystemTransientandHarmonicStudies UsingWaveletTransform",IEEETrans.PowerDelivery,vol.14,no.4,Oct.1999,pp.14611468. A.M.Gaouda,M.M.A.Salama,M.R.Sultan,andA.Y.Chikhani,"PowerQualityDetection and Classification Using Wavelet - Multiresolution Signal Decomposition", IEEE Trans. PowerDelivery,vol.14,no.4,Oct.1999,pp.1469-1476. L. Angrisani, P. Daponte, M. D'Apuzzo, and A.Testa, “A Measurement Method Based on WaveletTransformforPowerQualityAnalysis",IEEETrans.PowerDelivery,vol.13,no.4, Oct.1998,pp.990-998. C. DavidC.Robertson,OctaviaI.Camps,JeffreyS.Mayer,andWilliamB.Gish,“Wavelets andElectromagneticPowerSystemsTransients",IEEETrans.onPowerDelivery,vol.11, no.2,pp.1050-1057,April1996. [10] S. Santaso, E. J. Powers, W. M. Grady, and P. Hofmann, “Power Quality Assessment Via WaveletTransformAnalysis",IEEETrans.onPowerDelivery,vol.11,no.2,pp.924-930, April1996. [11] W.A.Wilkinson,andM.D.Cox,“DiscreteWaveletAnalysisofPowerSystemTransients", IEEETrans.onPowerSystems,vol.11,no.4,pp.2038-2044,November1996. [12] O.Caari,M.Meunier,andF.Brouaye,“Wavelets:AnewToolForTheResonantGrounded Power Distribution System Relaying", IEEE Trans. on Power Delivery, vol. 11, no. 3, pp. 1301-1308,July1996. [13] O.N.Gerek,M.N.Gurcan,A.E.Cetin,“FrequencyBandCharacteristicsofTree-Structured FilterBanks,”ElectronicsLetters,Vol.32,No.8,pp.724-726,11Apr.1996. [14] K.Ramchandran,M.Vetterli,andC.Herley,"Wavelets,SubbandCoding,andBestBases," ProceedingsoftheIEEE,Vol.84,No.4,pp.541-559,April1996. 5