搜索
http::(1983-),,;(),,,.:(50975213);(B08031).:2010-10-20:http:(,430063,):提出一种基于独立分量分析(ICA)和局部线性嵌入流形学习算法(LLE)的新型高压直流输电(VSC-HVDC)系统故障诊断方法.由于随机噪声的干扰,单个传感器测得的系统故障信号无法直接用于故障检测,故使用快速ICA对多通道传感器测得的直流电压和电流信号进行盲源分离处理以恢复去噪的系统故障源信号;然后利用LLE挖掘潜藏于恢复信号中的子流形,提取故障敏感特征;最后将LLE提取的故障特征量作为支持向量机(SVM)的输入,建立系统故障诊断模型.通过对系统交流相对相故障交流相对地故障以及复合故障等仿真信号进行分析,表明所提出的ICA-LLE方法能够有效地提取故障关键特征,并在3维空间将故障特征隔离,从而得到满意的SVM故障识别效果,且SVM分类精度比只使用LLE提高了近20%.:输电;故障诊断;独立分量分析;流形学习:TP273:A:0253-987X(2011)02-0044-05IndependentComponentAnalysisandManifoldLearningwithApplicationstoFaultDiagnosisofVSC-HVDCSystemsLIZhixiongYANXinping(SchoolofEnergyandPowerEngineering,WuhanUniversityofTechnology,Wuhan430063,China)Abstract:Afaultdiagnosisschemeforvoltagesourceconverter-highvoltagedirectcurrenttrans-mission(VSC-HVDC)systemisproposedbasedontheindependentcomponentanalysis(ICA)andthelocallylinearembedding(LLE)inthepresentwork.ThemeasuredsignalsintheVSC-HVDCcannotbeuseddirectlytodetectsystemfaultduetotheheavyinferencenoise.TheFastICAishenceemployedtoeliminatethedisturbednoiseandrecoverthefaultsourcesfromthemeasuredDClinevoltageandcurrentobservationsignals.ThentheLLEalgorithmisappliedtoextractdistinctcharacteristicshidingintherecoveredfaultsignals.Toenhancethefaultpatternrecognition,thesupportvectormachine(SVM)isadoptedtolearntherelationshipbetweenthefaultfeaturesandthesystemoperationconditions.TheabilityoftheproposedICA-LLEmethodtodetectVSC-HVDCsystemfaultisevaluatedwiththesimulateddata.Theanalysisresultsdemonstratethefeasibilityandeffectiveness.Thedistinguishedfeaturesofthefaultsignals,suchasACline-to-linefaultandACline-to-groundfault,andthecompoundfaults,canbeextractedefficientlyandthenisolatedinthe3-Dfeaturespacecorrectly.TheclassificationrateoftheSVMwiththeproposedschemeisincreasedby20%comparedwithLLE.Keywords:transmission;faultdiagnosis;independentcomponentanalysis;manifoldlearninghttp:(VSC-HVDC)[1-2],[2].,,,.,[3-7].,VSC-HVDC,[2].[3-5]HVDC,,,.,[6-7](SVM),SVM.,,,.VSC-HVDC,(ICA),ICA.(LLE),(SVM),.1(FastICA),.(ICA)[8-10],,,,.ICA[8]x=As(1):Ann;xm1;sn.ICAAW()^s=Wx(2):^ss.[8]ICA,Fas-tICA,10~100.FastICA,J(y)=[E(g(y))-E(g())]2(3):y=WTz,z;;g().W,(3)WW(k)=E{zg(WT(k-1)z)3}-E{g(WT(k-1)z)}W(k-1)(4):g()g().2(LLE),,.(Fisher),,[10].(LLE)[11],,.X={x1x2xn},xiRp(p),LLEXf()T,T={t1t2tn},tiRq(qp),ti=Tf(ti)X.LLE[11].(1).kxi.(2)V.Vmin(V)=ni=1xi-kj=1vijxij2(5):xij(j=1,2,,k)xi,vijxixij,kj=1vij=1.(5),uiVuijl=(xi-xij)T(xi-xil)(6),vij=kl=1(ui)-1jl/ka=1kb=1(ui)-1ab(7)(3).T452,:VSC-HVDChttp:(T)=ni=1nj=1mijtTitj=tr(TMTT)(8)M=(Inn-V)T(Inn-V)(9),TMq.,XT.LLE,SVM,.,,.1VSC-HVDC.1VSC-HVDC3VSC-HVDC31VSC-HVDC2.,VSC-HVDC,.[1],.Matlab/Simulink2,230kV,200MW,50Hz,100km.,,(LL)(TLG)(TLG-LG)(DS-LLL).2VSC-HVDC,,.3.,,.3VSC-HVDC32ICAVSC-HVDC,,,.ICA,,/10(10km),10.4FastICA10LL().,HVDC-19823dB,HVDC.4,FastICA3,LL.,,.,,,,ICA.,,(1)x1=x+dn=As+dn(10):dn.pICA,p=(11)5p.4645http:(-19823dB)5p67ICA(50dB).,,,.6ICA33LLESVMICA,LLE.LLE,d7ICAk,.dk.,SilhouetteLLE,.NCi(i=1,2,,N),a(x)x(xCi),d(x,Cj)xCj,b(x)=min{d(x,Cj)},j=1,2,,N,ji},xSilhouetteS=b(x)-a(x)max{b(x),a(x)}(12)Silhouette,,.Silhouette05,,05.8dk500(100)Silhouette.:234,k1~50(S05);d=2;k=35Silhouette05531(d=3)05649(d=4).d=3Silhouetted=4,d=33,k=35,d=3.9VSC-HVDCLLE.,,LLELLTLG-LG,TLGDS-LLL,(6),LLE,472,:VSC-HVDChttp:表1SVM诊断模型测试结果/%/%LLE-SVM36006006800800ICA-LLE-SVM38328366100.0100.04ICALLEVSC-HVDC,,SVM,.:20dB,FastICA;LLE,ICA-LLELLE;SVM.:[1]WEIMERSL.HVDClight:anewtechnologyforabetterenvironment[J].IEEEPowerEngineeringRe-view,1998,18(8):19-20.[2],,,.HVDC[J].,2008,28(1):23-29.YANBingyong,LIUXimei,TIANZuohua,etal.ApplicationofconsensusfilterandSVMinfaultdiag-nosisforHVDCsystems[J].ProceedingsoftheCSEE,2008,28(1):23-29.[3]NARENDRAKG,SOODVK,KHORASANIK,etal.Applicationofaradialbasisfunction(RBF)neuralnetworkforfaultdiagnosisinaHVDCsystem[J].IEEETransonPowerSystems,1998,13(1):177-183.[4]LAILL,NDEH-CHEF,CHARIT.HVDCsystemfaultdiagnosiswithneuralnetworks[C]FifthEuro-peanConferenceonPowerElectronicsandApplica-tions.Brighton,UK:[s.n.],1993:145-150.(58)4845http:(4):548-563.[5]XIAXG.Anewprefilterdesignfordiscretemult-iwavelettransforms[J].IEEETransactionsonSignalProcessing,1998,46(6):1558-1570.[6]CHEUNGKW,POLM.Integermultiwavelettrans-formforlosslessimagecoding[C]Proceedingsof2001InternationalSymposiumonIntelligentMultime-diaVideoandSpeechProcessing.Piscataway,NJ,USA:IEEE,2001:117-120.[7]HUANGB,SRIRAJAY,HUANGHL,etal.Loss-lessmultiwaveletcompressionofultraspectralsounderdata[C]IEEEInternationalConferenceonGeosc-ienceandRemoteSensingSymposium.Piscataway,NJ,USA:IEEE,2006:3541-3544.[8]VANFLEETPJ.Multiwaveletsandintegertrans-forms[J].JournalofComputationalAnalysisandAp-plications,2003,5(1):161-178.[9]JINGMingli,HUANGHua,LIUWuling,etal.Ageneralapproachfororthogonal