Seite - 367 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1253 Figure10.REofpredictionmethods. Thestatisticalerrorsof thefivepredictionmodelsaredisplayedinFigure11. Theanalysis shows that: (a)NILA-CNNmodel outperformsother four techniques in termsofRMSE (2.27%),MAPE (2.14%)andAAE (2.096%). (b)ComparedwithLA-CNN,NIavoidsprematureconvergencebasedon increasingthediversityof lionpopulation. (c)Thegeneralizationabilityandpredictionaccuracyof theCNNmodelcanbe improvedbyparameteroptimization. (d) theCNNmodelcanmakeadeep excavationof the internal relationshipbetween the influential factors and the loadofEVcharging station incomparisonwithSVM. (e)ANNcanreflect thenon-linear relationshipmoreaccurately than TSmethods. Figure11.RMSE,MAPEandAAEofpredictionmethods (I). 5. FurtherStudy Inorder to furtherverify theeffectivenessof theproposedmodel,onemorecasewhichselects the data fromanotherEVchargingstation isprovidedin thispaper. Thestudyiscarriedoutwithdata from1June2016 to31May2017. Toreflect the influenceofseasonal factorson load,data from7days of eachseasonare selectedasa test set,with the rest asa trainingset. Thespecificdatadivision is showninTable3. 367