Optimization of support vector machine power load,,,,

《Optimization of support vector machine power load,,,,》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、J.Cent.SouthUniv.Technol.(2010)17:406−412DOI:10.1007/s11771−010−0060−0OptimizationofsupportvectormachinepowerloadforecastingmodelbasedondataminingandLyapunovexponentsNIUDong-xiao(牛东晓),WANGYong-li(王永利),MAXiao-yong(马小勇)SchoolofEconomicsandManagement,NorthChinaElectric

2、PowerUniversity,Beijing102206,China©CentralSouthUniversityPressandSpringer-VerlagBerlinHeidelberg2010Abstract:Accordingtothechaoticandnon-linearcharactersofpowerloaddata,thetimeseriesmatrixisestablishedwiththetheoryofphase-spacereconstruction,andthenLyapunovexponent

3、swithchaotictimeseriesarecomputedtodeterminethetimedelayandtheembeddingdimension.Duetodifferentfeaturesofthedata,dataminingalgorithmisconductedtoclassifythedataintodifferentgroups.Redundantinformationiseliminatedbytheadvantageofdataminingtechnology,andthehistoricall

4、oadsthathavehighlysimilarfeatureswiththeforecastingdayaresearchedbythesystem.Asaresult,thetrainingdatacanbedecreasedandthecomputingspeedcanalsobeimprovedwhenconstructingsupportvectormachine(SVM)model.Then,SVMalgorithmisusedtopredictpowerloadwithparametersthatgetinpr

5、etreatment.Inordertoprovetheeffectivenessofthenewmodel,thecalculationwithdataminingSVMalgorithmiscomparedwiththatofsingleSVMandbackpropagationnetwork.ItcanbeseenthatthenewDSVMalgorithmeffectivelyimprovestheforecastaccuracyby0.75%,1.10%and1.73%comparedwithSVMfortwora

6、ndomdimensionsof11-dimension,14-dimensionandBPnetwork,respectively.ThisindicatesthattheDSVMgainsperfectimprovementeffectintheshort-termpowerloadforecasting.Keywords:powerloadforecasting;supportvectormachine(SVM);Lyapunovexponent;datamining;embeddingdimension;feature

7、classificationAccordingtothechaoticsolutionwithstochasticnature1Introductionintheinherentcertaintyofnonlinearsystems,itispredictableintheshorttermbutunpredictableinthelongDuringrecentdecades,numerousinvestigationsterm.Atpresentchaotictimeseriesforecastingmethodshave

8、beencarriedouttoimprovetheaccuracyofincludeglobalprediction[10],andforecastmethodelectricityloadforecasting.OnemethodistoforecastbasedonLy