methodsasssd

PredictionofElectricLoadNeuralNetworkPredictionModelforBigDataGuochenJin1,*,XiangyingTang2,DepingMiao21Departmentofxxxx,yyyyUniversity,Beijing,China2Schoolofaaaa,bbbbUniversity,Changsha,China*Correspondingauthor:cccc@dddd.
comKeywords:NeuralNetwork,PredictionModel,BigData.
Abstract:Powerloadforecastingisveryimportantforpowerdispatching.
Accurateloadforecastingisofgreatsignificanceforsavingenergy,reducinggeneratingcostandimprovingsocialandeconomicbenefits.
Inordertoaccuratelypredictthepowerload,basedonBPneuralnetworktheory,combinedwiththeadvantagesofClementineindealingwithbigdataandpreventingoverfitting,aneuralnetworkpredictionmodelforlargedataisconstructed.
IntroductionTheaccuratepredictionofpowerloadisofgreatsignificancefortheelectricpowerproductionandthesafeoperationofthepowergridandthenationaleconomy[1].
Shorttermloadforecastingisanimportantpartofenergymanagementsystem.
Thepredictionerrordirectlyaffectstheanalysisresultsofsubsequentsafetycheckofpowergrid,whichisofgreatsignificancefordynamicstateestimation,loadschedulingandcostreduction[2-4].
Traditionalpredictionmethodsarebasedonlinearregression,suchastimeseriesmethod,analysismethodandpatternrecognitionmethodhasdefectsofrespectively[5].
ThebasicfunamentalofBPneuralnetwork2.
1ThestructureofBPneuralnetworkBPneuralnetworkisamulti-layernetworkwitherrorreversepropagation,whichiscomposedofinputlayernodes,hiddenlayernodesandoutputlayernodes.
Thisprocesshasbeenreducedtoanacceptableleveloferrortothenetworkoutput,ortoapredeterminednumberoflearningtimes.
ThenetworkstructureisshowninFigure1.
Figure1.
NeuralnetworkstructureThegeneralmodelofartificialneuralnetworkconsistsoffourbasicelements,whichare:(1)TheBPneuralnetworkislinkedbydifferentnodecoefficients.
Whenconnectingweightsandweightsarepositive,itindicatesthatthecurrentlinkisanexcitingstate.
Conversely,ifthelinkcoefficientisnegative,thelinkstateisastateofsuppression.
(2)Theinputsignalandthelinearsignalarethecombinationofthesignalsforeachinputsignal.
(3)Thefunctionofthenonlinearactivationfunction:makingtheneuronoutputsignalwithinacertainrange.
(1)(2)(3)BPneuralnetworkisbackpropagating,mainlycomposedofthreeparts:inputlayer,middlelayerandoutputlayer.
Thenumberofnodesintheinputandoutputlayersisrelativelyeasytodetermine,butthedeterminationofthenumberofnodesinthehiddenlayerisaveryimportantandcomplexproblem.
2.
2ThedeterminationofthenumberofnetworklayersBPneuralnetworkisbackpropagating,mainlycomposedofthreeparts:inputlayer,middlelayerandoutputlayer.
Thenumberofnodesintheinputandoutputlayersisrelativelyeasytodetermine,butthedeterminationofthenumberofnodesinthehiddenlayerisaveryimportantandcomplexproblem.
Results3.
1TheestablishmentofsimulationmodelThelargedatapredictionmodelfortheuser'selectricityconsumptionisimplementedintheClementinesoftware.
3.
2AnalysisofexperimentalresultsByselectingtheloadpredictionresultsof403and411lines.
Wecanseethattheactualvaluesofthelinesbasicallymatchthepredictedvalues,buttherearealsosomeerrors,especiallyinthepeakperiodofelectricityconsumption,asshowninTable.
1.
Table.
1.
Comparisonofpowerloadforecastingof403lineComparisonPowerForecastingA1293792387B92873529837C89452323894Fromthecomparisonbetweenpredictiondataandactualdata,theBPneuralnetworkhasbetterpredictionperformanceandrelativelysmallerror,whichcanmeetthedemandcompletely,andhasfastpredictionspeedandconvenientoperation.
ConclusionsThetrendofmassdatainpowersystemprovidesabasisforloadcharacteristicanalysisandpredictionmodelestablishment,buttheclassicalloadforecastingmethodcannotaffordsuchahugetimeandcomputingresourceconsumption.
Theproblemofoverfittinginlargesamplesetwillaffectthepredictionaccuracy.
Inthispaper,apowerloadforecastingmodelisbuiltbyusingtheBPneuralnetworkmodel,makingfulluseofthepowerfuldataprocessingfunctionofClementineandpreventingtheoverfittingfunction.
TheexperimentalresultsshowthattheBPneuralnetworkmodelhasgoodpredictabilityandrobustness,andhasacertainpracticalapplicationvalue.
AcknowledgementsTheauthorsgratefullyacknowledgethefinancialsupportfromxxxfunds.
ReferencesChengQiyun,SunCaixin,ZhangXiaoxing,etal.
Short-Termloadforecastingmodelandmethodforpowersystembasedoncomplementationofneuralnetworkandfuzzylogic[J].
TransactionsofChinaElectrotechnicalSociety,2004,19(10):53-58.
Fangfang.
ResearchonpowerloadforecastingbasedonImprovedBPneuralnetwork[D].
HarbinInstituteofTechnology,2011.
AmjadyN.
Short-termhourlyloadforecastingusingtimeseriesmodelingwithpeakloadestimationcapability[J].
IEEETransactionsonPowerSystems,2001,16(4):798-805.
MaKunlong.
Shorttermdistributedloadforecastingmethodbasedonbigdata[D].
Changsha:HunanUniversity,2014.
SHIBiao,LIYuXia,YUXhua,YANWang.
Short-termloadforecastingbasedonmodifiedparticleswarmoptimizerandfuzzyneuralnetworkmodel[J].
SystemsEngineering-TheoryandPractice,2010,30(1):158-160.