上海交通大学硕士学位论文考虑气象要素的电力系统短期负荷预测研究姓名:王波申请学位级别:硕士专业:电力系统及其自动化指导教师:邰能灵20080101I(CPSO_ARMAX)IIABSTRACTIIITEMPERATURESENSITIVEMETHODFORSHORT-TERMLOADFORECASTINGABSTRACTShort-TermLoadForecasting(STLF)playsanimportantroleinpowersystemplanningandoperation.Basicoperationfunctionssuchasunitcommitment,economicdispatch,fuelscheduling,andunitmaintenancecanbeperformedmoreefficientlywithanaccurateforecasting.However,theSTLFisdifficulttohandleduetothenonlinearandrandom-likebehaviorsofsystemload,weatherconditions,andvariationsofsocialandeconomicenvironments,etc.Amongtheclassicalapproaches,theStochasticTimeSeriestechniqueseemstobethemostpopularmethod.Theadvantageofthissortofmethodisthatthemodelwithminimummeansquareforecastingerrorissearched.However,thetraditionaltimeseriesforecastingmodelcannothandletheweatherinfluences,itcannotreachtheprecisionrequirementofmodernloadforecasting.Typically,theloadpatternofpowersystemspossessesnonlinearanddynamiccharacteristics.Inadditiontoseasonalanddiurnalvariation,theloadshapeisgenerallygovernedbytheambientweatherconditions.Inordertocapturetherelationshipbetweentheloadandtheweatheraccurately,anARMAXmodelhasbeendevelopedsuccessfully.Sincethemeansquareforecastingerrorfunctionpossessesmultiplelocalminimumpoints,thezerogradientpointcannotbeguaranteedastheglobaloptimalsolution.Thetraditionalgradient-basedidentificationtechniqueusedbythetraditionalapproachispronetostallatthelocalminimumpoint.ThispaperproposesanewparticleswarmABSTRACTIVoptimizationbasedonchaoticmappinginertiaweighttoidentifyARMAXmodelforSTLF.SincetheCPSOevaluatessimultaneouslymanypointsinthesearchspace,itiscapableofasymptoticallycoveringtowardtheglobalminimumsolutionandcanthusimprovethefittingaccuracyofthemodel.Thismethodisstableandrobust,ittakesadvantageofchaoticmappingtoenhancetheglobalsearchabilityofPSO.TestingonthepracticalloaddataofShanghaipowermarketforone-dayaheadshort-termloadforecastinghasshownthattheproposedmethodhasgoodaccuracy.Andthewholeprocessrequiresnoexpertiseormanualintervention,sothismethodisofmuchapplicability.KEYWORDSShort-TermLoadForecasting,Chaotic,ParticleSwarmOptimization,Auto-RegressiveandMovingAveragewithexogenousvariables200812120081212008121111.1115AGC5(1)2(2)[1-2](3)(4)(5)31.21.2.1(1)(2)1()()niiiytabxt==+×∑(1.1)n()ixtt1,,,nabbL(3)4ARMAARMAARIMA1.2.22080(1)BP[3-5]RBF[6-7]Elman[8-9]Hopfield[10](2)VapnikSVMVC(Vapnikchervoneks)[11][11-13][14],5(3)[15]1.2.3[16][17-18]ARMA[19]1.3(1)[20][21]6-[22-23]ARMAX(2)[24]24[25][26][27][28]SVM[29][30]SVMs[31]41.47(1)SCADA(2)(3)mm(4)1.5(1)(2)(3)ARMAX(4)ARMAX8CPSO_ARMAX922.1:(1)(2)GDP(3)1020032006200320060.82.22.2.120032006(1)177116112369911231105010015020025030035040050006000700080009000100001100012000T/Load/MW(a)200312050100150200250300350400600070008000900010000110001200013000T/Load/MW(b)20040501001502002503003504000.70.80.911.11.21.31.41.5x104T/Load/MW(c)2005130501001502002503003504000.60.811.21.41.61.8x104T/Load/MW(d)20061Fig1.Thecarveofmeanvaluefordiurnalload,2003-20062.2.224811681401020304050607080901000.80.850.90.9511.051.11.151.21.251.3x104T/Load/MW(a)49601020304050607080901000.911.11.21.31.41.51.6x104T/Load/MW(b)7961501020304050607080901007000750080008500900095001000010500110001150012000T/Load/MW(c)1196296Fig2.The96-pointloadcarveoftypicalday2.2.31.320054120054796162.44200546500700075008000850090009500100001050011000LOAD/MW341796Fig3.The96-pointloadcarveof4/1/2005to4/7/2005174341041442180008200840086008800900092009400LOAD/MW44321Fig4.Thecarveofmeanvaluefordiurnalloadfrom4/3/2005to4/21/20053.1120032006781.212()4.182.32090200296.8SPSS14.0(StatisticalProgramforSocialSciences,SPSS)12006Table1.Thecorrelationanalysisbetweenwinterloadandweather,20061160.833-0.040.359-0.7560.870.8760.4780.1031230.510-0.2050.3310.4180.5410.619-0.711300.3580.0450.4570.3750.4430.204-0.46326-0.260.1180.056-0.381-0.33-0.380.575-0.9092130.797-0.1630.3480.2040.7070.796-0.674-0.6191922005Table2.Thecorrelationanalysisbetweenspringloadandweather,20053160.7540.169-0.3810.2320.6970.733-0.5590.5163230.8900.169-0.095-0.7350.8450.8920.414-0.8133300.8330.1610.5030.7590.836-0.646-0.693460.538-0.298-0.033-0.2080.4160.541-0.6300.0224130.778-0.2370.5970.7480.8100.537-0.80732005Table3.Thecorrelationanalysisbetweenautumnloadandweather,200510100.844-0.0550.5420.8080.853-0.178-0.79510170.91-0.0590.3710.8430.9050.581-0.84510240.888-0.310.3470.820.883-0.162-0.80210310.831-0.0010.3030.76710.822-0.156-0.7121170.6930.260.210.6330.712-0.366-0.57942005Table4.Thecorrelationanalysisbetweensummerloadandweather,200576-0.0030.0380.097-0.237-0.089-0.0610.373-0.377130.875-0.1080.2570.5260.8650.8930.677-0.7837200.871-0.2280.062-0.3580.8840.853-0.626-0.8717270.919-0.1210.1460.8850.920-0.638-0.894830.8460.2980.4940.8280.8680.056-0.941762025-26200572072969780.835oC35oC40oC38oC510(1)1078(2)5oC7oC.0.80.850.7()10467892.2152006Table5.Thepartialcorrelationanalysisofwinterloadandweather,2006116-0.4290.407-0.7880.4890.4550.733-0.526123-0.0970.307-0.2350.2400.878-0.706130-0.0020.4760.1620.5580.285-0.388260.0890.095-0.304-0.439-0.5710.532