Seite - 360 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1253 beguaranteed,dueto thesubjectivedeterminationof theweightsandthresholds [26]; thus,NILAis proposedtocomplete theoptimalparameterselection in thispaper toovercomethisshortcoming. 2.3. TheForecastingModel ofNILA-CNN Theshort-termloadforecastingapproachforEVchargingstations incorporatingNILAandCNN isconstructedasFigure3shows. Original load data Data pretreatment Training set Testing set Initialized the parameters of NILA Population initialization Mating Territorial defense Territorial takeover Lion clone Single parent mutation Stop criteria Obtain the optimal parameters CNN Output the forecasting results YesNo Niche immune Lion algorithm improved by niche immunity Figure3.FlowchartofLionAlgorithmImprovedbyNicheImmune(NILA)-ConvolutionalNeural Network(CNN)algorithm. OnthebasisofNILA-CNNmodel, theoptimalparametersofCNNcanbederivedas follows: (1) Input selection (xi) anddata pre-processing. The initial input set is formedbased on the loadanalysisofEVchargingstationsandneeds tobequantifiedandnormalized. Thespecificdata preprocessingmethodisshowninSection4.1. (2)Parameters initialization. Randomlydetermine theweights and thresholdsof all layers in CNNmodel fromthesmallernumerical set. 360