基于负荷特性的短期电力负荷预测系统开发

上海交通大学硕士学位论文基于负荷特性的短期电力负荷预测系统开发姓名：郑健申请学位级别：硕士专业：电力系统及其自动化指导教师：程浩忠;杨健20060201I,2410kV35kV110kVIIIIIABSTRACTElectricshort-termloadforecasting(STLF)isanimportantcomponentinthedailyoperationoftheelectricutility.TheSTLFcanpredictelectricloadforaperiodofhours,days,orweeks,especiallyforthenexttwenty-fourhours,whichistheprimarybasistoestablishthepowergenerationplanandthetransmissionscheme.Itisdifficulttoforecasttheshort-termloadquantitybecausethepredictioncanbeinfluencedbymanyfactors.Therefore,itwillbeofrealsignificanceintheincreaseofforecastaccuracytoputtheexistingforecastmodelstogooduseandstudythenewoneswhichconsideringtheloadcharacteristicsatthesametime.SeveralpracticalSTLFmodels,suchastimeseries,graymodel,leastmeansquares,andsimilardayarestudiedinthispaper;integratedoptimummodelisstudiedforbetterutilizationoftheresultsofmultiplemodels,whichismoresuitableinSTLF.ThepeculiaritiesoftheloadssuppliedbytheHuxiPowerSupplyCompanyarestudiedindetails.Thedailyloadcharacteristicsandthecurvesofthevarietytrendarepresented.TheSTLFmodelfortheworkdaywiththemaximalloadquantitytakenintoaccountandtheonefortheholidayareputforwardrespectively.Intheoperationworkofthepowersystem,thedispatchershouldknownotonlythetotalloadquantitybutalsotheloadofthedifferentIVvoltageclassestoestimatetherunningconditionsofthedevices.Atthesametimetheforecastdatumondifferentvoltagelevelscanbeusedasthereferencebasisfortherecoveryservice.AwholeSTLFsystembasedonDispatchingAutomationSystemfordistrictpowernetworksissuccessfullydevelopedaccordingtothepracticaldemandsofelectricdepartmentaswellastheoperationalguidanceconditionsandtheexistingstateofthepowersystem,whichisadvantageofreal-time,economyandpracticality.Theloadquantityof10kV,35kV,110kVandthetotalpowerloadcanbeforecastedintheSLTFsystemrespectivelyasrequired.TheSTLFsystemalsoprovidesadata-platformforthedispatchers.TheloadcharacteristicsandthemeteorologicalphenomenaaretakenintoaccountintheSTLFsystem,whichcanimplementsomeexpandedfunctionsbyaddingprogrammodules.Ithasbeenprovedbypracticaldatathatthesystemcancommendablysatisfydemandsofshorttermloadforecastingandpresentaccuratefutureloadmagnitudeswithfriendlyman-machineinterfacesandconvenientaccessesfunctions.KEYWORDS:powersystem,powerload,short-termloadforecasting(STLF),loadcharacteristics,forecastingmodel,systemdeveloping11.1[1,2]EMS[2][2]1.2EMS2[1,2]1[1][2]ARIMA[1]GM11GM1n3[2,3]ANNBPRBFHopfield[49][4][5][6][10][11][12][28]ARIMA[11]RadialBaseFunctionRBF[13]3244S-BP[1416][21,22][23]RBFEA5[24,25]Fourier[20,21]2[26][27]—15-2036Bayesian[28]ARBayesian[1]1.3961123212SCADA73482.112aIf|(,)(,1)}|()|(,)(,1)}|()LdtLdttLdtLdttαβ−−−+Then(,)((,1)(,1))/2LdtLdtLdt=−++(2-1)(,)Ldtdt()tα()tβb249If|(,)()|()LdtLttθ−Then()()(,)()(,)()()(,)()LttLdtLtLdtLttLdtLtθθ+=−(2-2)()Ltt11()(,)ndLtLdtn==∑(2-3)()tθc1122(,)(,)(,)LdtLdtLdtωω=+(2-4)1(,)Ldt2(,)Ldtd1ω2ω3410a(AbsoluteError)ˆ()|()()|AEtLtLt=−(2-5)bRelativeErrorˆ|()()|()()100%100%()()LtLtAEtREtLtLt−=×=×(2-6)cMeanAbsoluteError11()()niiMAEtAEtn==∑(2-7)d242421()100%24iidREtE==×∑(2-8)1ddAE=−(2-9)ˆ()Ltt()Ltt2.2()()()()()LtBtWtStVt=+++(2-10)2-10()Lt()Bt()Wt()St()Vt()Bt()Wt11()St()Vt122490BPRBFHopfieldElman3.13.1.113011()()...()()nnytbbxtbxttθ=++++(3-1)()ytt()(1,2,...,)ixtin=(0,1,...,)jbjn=()tθ2(0,)Nσ3-1jb3.1.2143.1.3,,,,[1]()yt()tθ3-1ARMAARMAARIMASeasonalProcessTFARMA3-1Fig3-1TimeSequenceModelforLoadARMA()ytARMA1212()(1)(2)...()()(1)(2)...()pqytytytytpttttqφφφθωθωθωθ=−+−++−+−−−−−−−(3-2)()ytt()tθtpqiφiωARMApq()tθ()yt15p12,,...,pφφφq12,,...,qωωω3.1.4a1212(,)(,,...,)(,,...,)fhTnTnPPPabaaaabbbb∆=−=Φ==P∆fPhP12,,,naaa12,,,nbbb,abab−ab−112233***abKKKδδδ−=++161K2K3Kab−A1δ1,1,DDDLNDDLδ∆∆≤=∆D∆NDLB2δC3δ1γ2γ3γ3112233cccδγγγ=++123,,cccb173.23.2.140W.S.McCullochW.Pitts80D.E.RumelhartBPNNBack-PropagationNeuralNetwork13-2181x2x•••1y2y••••••MyLx2z1z3zHz3-2Fig3-2Three-layerForwardNeuralNetworkRumeejartMcCllandBPBP3-212(,,,)Lxxx12(,,,)Hsss12(,,,)Hzzz12(,,,)Myyy01011()11LiijjijiiHkkjjkjswxwiHzsiHyvzvkMσ===+≤≤=≤≤=+≤≤∑∑(3-3)ijw-0iwkiv-0kvMkvwxwvyHikLjijijkik≤≤++=∑∑==1)(1010σ(3-4)19BPwijvkiBPBPBP21985PowellRBF1988BroomheadLowe1x2xnx••1y2yny••••()iRxikw3-3RBFFig3-3RBFNeuralNetwork20RBFRBFRBFRBFRBF2||||()exp[]1,2,...,2iiixcRximσ−=−=(3-5)xnicixiσim1()()1,2,...,mikiiyfxwRxkp====∑(3-6)pRBFBP:BP,,,,BP,RBFBP,BPSigmoid,,RBF3214HopfieldHopfield1982HopfieldElmanElmanElman3.2.2122XX1XX={x}XA:[0,1]Axµ−AµA()AxµxAx()Axµ1xAx2ZadehABXABxX∈()()()max{(),()}ABABABxxxxxµµµµµ∪=∨=(3-7)ABxX∈()()()min{(),()}ABABABxxxxxµµµµµ∩=∧=(3-8)ABxX∈()1()AAxxµµ=−(3-9)3XYRXYR,()ijRr=(3-10)4xAxB()BxµxA()AxµxB()()ABxxµµ∧2BP233.2.312243.3BPRBFHopfieldElmanBPBP25BPBP264.1274.25()50.244.2.1200467200461913288134.102004006008001000135270310541405175621072458280931603511(MW)4.1Fig4.1theloadcurveofcontinuous13days284.2.2200467200412134.2780200400600800100012001357911131517192123252729(2004.6.7-2004.12.13)MW4.2Fig4.2thecurveofthemaximumloadinworkdays4.2.3200465200412184.3290100200300400500600700800900100013579111315171921232527292004.6.52004.12.18MW4.3Fig4.3thecurveofthemaximumloadinweek