搜索
268Vol.26No.8Apr.200620064ProceedingsoftheCSEE©2006Chin.Soc.forElec.Eng.0258-8013(2006)08-0117-07TP391.4A470⋅401123(141008224100073410012)ContaminationGradesRecognitionofInsulatorsUnderDifferentHumidityUsingInfraredImageFeaturesandRBPNNHEHong-ying1,YAOJian-gang1,JIANGZheng-long2,LIWei-wei3(1.SchoolofElectrical&InformationEngineering,HunanUniversity,Changsha410082,HunanProvince,China;2.HunanProvinceElectricPowerTestAcademy,Changsha410007,HunanProvince,China;3.HunanHDHLElectrical&InformationTechco.,LTD,Changsha410012,HunanProvince,China)ABSTRACT:Anewmethodispresented,usinginfraredimagefeaturesandradialbasisprobabilisticneuralnetwork(RBPNN)forinsulatorcontaminationgradesdetectionofnaturalcontaminatedinsulatorsunderdifferenthumidity.Anamendedalphafilterandanimagesegmentationmethodbasedonthehistogramtroughoftheinsulatorimageareadoptedtopreprocesstheinsulatorinfraredimage.Experimentsaredesignedunderdifferenthumidity,fourfeatureswhichcanrepresentthecontaminationgradesoftheinsulatorareextractedfromthesegmentedimage,theyarethemeantemperatureoftheimagebackground,thehighesttemperatureoftheinsulatorsurface,themeantemperatureoftheinsulatorsurface,thetemperaturevarianceoftheinsulatorsurface,RBPNNisdesignedtomaptherelationbetweentheinfraredimagefeaturesoftheinsulatorunderdifferenthumidityandthecontaminationgrades;thenRBPNNistrainedtorecognizethecontaminationgrades.AmethodintegratedgradientalgorithmandrandomalgorithmisappliedtodecidethecentersofhiddennodesthewidthcontrolparametersandtheweightsmatrixoftheRBPNN.Experimentsresultsindicatethismethodisaneffectiveapproachforthedetectionoftheinsulatorcontaminationgrades.KEYWORDS:contaminatedinsulatorinfraredimagefeatures;amendedalphafilter;imagesegmentation;radialbasisprobabi-([2002]845)([2003]790)([2003]005)listicneuralnetwork;gradientalgorithmandrandomalgorithm;contaminationgradesrecognition(RBPNN)()4RBPNNRBPNNRBPNN0[1-3][4-5][6][7-8][9][10-12]11826PC(RBPNN)41[13]FLIRPM69522.1PM695IMG()IRWIN5.31(RBPNN)RBPNNRBPNNRBPNN1RBPNN1Fig.1Flowofalgorithm2.2[14]m×nSxy(x,y)(s,t)f(s,t)d/2d/2fr(s,t)mn-df(x,y)ˆ(,)fxy(,)1ˆ(,)(,)xyrstfxyfstmnds∈=-∑(1)Sxymn=5×5d=62.38119()2Tf(x,y)TT0032Fig.2Thehistogramofinsulatorimage3Fig.3Segmentedimageofinsulator1ini1LiiNn==∑(2)()inPiN=(3)i=1,…,L23()4T5T(,),(,)(,)0,(,)fxyfxyTfxyfxyT≥⎧⎪=⎨⎪⎩(4)f(x,y)2.4()()(Tcmean)(Tmax)(Tmean)(Evcar)45RBPNN2.52.5.1(RadialbasisprobabilisticneuralnetworksRBPNN)[15](RBFNN)(PNN)()PNN[16]RBPNN4()PRRBPNNRMM4Fig.4Structureoftheradialbasisprobabilisticneuralnetwork12026()M0012NRBPNN(5)1112121222121112112......[][]...............NNNMMMMNhhhhhhYYY⎡⎤⎢⎥⎢⎥=⎢⎥⎢⎥⎣⎦KK(5)Y=WH(6)221()iqkijtjtkChFs=-=-∑P(7)Y1,Y2,,YNNWHP(R)F(⋅)()qii()CkkP⋅22skk2.5.2CkskWRBFNNRBPNNK-Mmean[17][18]EMK-MmeanK-Mmean[19]CkskWCkskW1dkdk0N(0,dk)2Cksk(8)(9)Ck(t+1)=Ck(t)+N(0,dk(t))Ck(t)(8)sk(t+1)=sk(t)+N(0,dk(t))sk(t)(9)Ckskkt3Ck(t+1)sk(t+1)t+1211(1)[()()]NjjtjetyPfP+=+=-∑(10)Ck(t)sk(t)t21()[()()]NjjtjetyPfP==-∑(11)Pjjyt(Pj)Pjtyt+1(Pj)Pjt+1f(Pj)Pje(t+1)e(t)Ck(t+1)sk(t+1)(8)(9)Ck(t+1)=Ck(t)sk(t+1)=sk(t)4dk(12)1122(),(1)(),1(1)(),(1)(),1kkkbtetetbtbtetetbddd+≤⎧+=⎨+⎩(12)b11.18b20.865(13)(())(1)()()iiietwtwtwth∂+=-∂(13)wi(t)i8121h0.326(1)~(5)e(t+1)e33.1050(10)XP-70501050HDDG-9533BM/50~0.03mg/cm20.03~0.06mg/cm20.06~0.10mg/cm20.10~0.25mg/cm20.25~0.350(GB/TKGO5582-93)(3m×3m×3m)XP-70(0~)110kV()XP-707XP-70110/(37)⋅≈10kVXP-7010kV5R20K()WSU7150L0409()KZX-51022JZYD10kV5h()PM6953m6(5)3010RWRWRWRWXP-70JZ5kVAKZX-51022RW5Fig.5Simplifiedcircuitmapofexperiment12001000RBPNN200688%88%6Fig.6Originalimageofinsulatorsonallcontaminationgrades3.2(RBPNN)RBPNN554850.321000RBPNN21600.0308%100%91.12%1(83%88%93%96%4)1~2021~25Tcmean()TmaxTmeanEvcar451RBPNN122261Tab.1Partsamplesfortrainingortestingandoutputresultsofexperiment/%Tcmean/Tmax/Tcmean/Evcar/F18331.21732.42531.5910.0260028331.36334.21032.0710.21738331.38535.84432.6590.27648331.21837.09233.1180.39958331.15640.58333.8260.87368831.24132.46531.6780.0270078830.64735.97432.2660.41988830.73938.40332.4970.74398830.78840.21033.0890.999108831.03342.87234.2091.637119330.82232.34131.5970.03900129330.93738.80133.0980.659139330.29641.75033.5182.272149330.21146.99333.8593.235159330.22352.92034.6684.925169630.31232.17731.2890.04100179630.51541.84232.9741.617189630.19644.75033.5182.872199630.12949.25734.0653.983209630.04856.17134.4335.805*218331.25932.61831.6590.03000*228830.54536.19932.3160.445*239330.28242.14433.6532.288*249330.35547.38333.9113.311*259630.31157.12934.9915.961*F11234RBPNN5RBPNN(5)(0~)4()4[1][J]200222(11)115-120SunCaixinSHULichunJiangXingliangetalAC/DCflashoverperformanceanditsvotagecorrectionofUHVinsulatorsinhighaltitudeandicingpollutionenvironments[J]ProceedingsoftheCSEE200222(11)115-120(inChinese)[2][J]200323(3)131-136NieYixiongYinXianggenLiuChunetalEvalutionontheon-linedetectionresultsofinsulatorstringsusingfuzzylogicmethod[J]ProceedingsoftheCSEE200323(3)131-136(inChinese)[3]750kV[J]200525(12)159-164JiangXingliangZhangZhijinHuJianlinetalACpollutionflashoverperformanceandcomparisonofshortsamples0f750kvcompositeinsulatorswithdiffer