Seite - 332 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1449 optimization in LSSVMmodel. In comparisonwith BPNN, LSSVMcan avoid the drawbacks of prematureconvergenceandeasily falling into localoptimum. Figure11.RMSE,MAPEandAAEof the forecastingresults (I). 5. FurtherStudy Inorder to furtherverify thevalidityof theproposedmethod,anothere-buschargingstation in Baoding,China,wasselected foranexperimental study. The loaddataof thestation fromJanuary, 2016 toDecember,2016areprovidedin thispaper,wheresevensuccessivedays ineachseasonwere takenas test samplesandtheremainingdatawereusedas trainingsamples. Thesettingofparameters inWPA-LSSVMwasconsistentwith theproposedmethod. InLSSVM,γandσ2wereequal to10.2801 and11.2675, respectively. Thevaluesof theparameters in theBPNNmodelweresameas those in the previouscasestudy. Figure12displays thevaluesofRMSE,MAPEandAAE. Figure 12. RMSE,MAPEandAAEof the forecasting results (II): (a) Spring test; (b) Summer test; (c)Autumntest; (d)Winter test. 332