Seite - 356 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1253 high.Also, it cannot fullymeet theneedsof theCNNparameteroptimizationproblem[21]. TheLion algorithm(LA),basedonthesocialbehaviorof lions,was introducedbyB.R.Rajakumar in2012 [22]. Comparedwithprecedingmodels, thisapproachshowsstrongrobustnessandgoodabilities inglobal optimization, and fast convergence. Nevertheless, inbreedingappears among the lionswith large fitnessduringthe iterativeprocess,which leads toprematureconvergenceanddiversityreduction. Tosettle thisproblem,niche immunealgorithmsareemployedinthispaper tooptimizeLA,namely NILA.Here,niche immunealgorithmscanrestrictover-duplicationof similar individuals, soas to ensure thediversityof thepopulation,andimprovetheoptimizationeffectof the lionalgorithmfor selecting theparametersofCNN.Thishybridoptimizationmethodisusedtoautomaticallydetermine theappropriatevalues inCNNmodel. Thispaper combinesNILAwith theCNNmodel for load forecastingofEVchargingstations, withscientificanalysisof influential factors. Therestof thepaper isorganizedas follows: Section2 showsabriefdescriptionofLA,NILA,andCNN,aswellas the frameworkof theproposedtechnique; Section3presentsananalysisof the influential factorsanddetermines the input;Section4 introduces anexperimentstudyto test theaccuracyandrobustnessof theestablishedmodel; Section5makes furthervalidationonthismethod,andSection6concludes thispaper. The innovationsof thispaperareas follows: (1) Theconstructionof the forecastingmodel Firstly, it is thefirst timetocombineCNNandlionalgorithmimprovedbyniche immunityand employthismodel for the loadforecastingofelectricvehiclechargingstations. Furthermore, theCNN modelusedfor loadforecastingcannotonlyallowtheexistenceofdeformeddata,butalso improve the load forecastingefficiencyandaccuracybyparameter reduction through local connectionand sharedweight. Finally, niche immunity isused in thispaper to restrict overduplicationof similar individuals, soas toensure thediversityofpopulation,anditeffectively improves theoptimization effectof the lionalgorithm,aswecanconcludefromthecasestudy. (2) The inputselectionof the forecasting Inorder toproduce a scientific and reasonable input index system for the forecastingmodel, thispaper fullyanalyzes the loadcharacteristics inanEVchargingstation.Anditcanbeconcluded that the loadin theEVchargingstation isheavily influencedbymeteorological conditions, seasonal variation,anddaytypes,whicharemorecomprehensiveandeffective for forecasting. Insummary, thispapernotonlycreativelycombinesvariouspredictiontheories toconstructa comprehensive forecastingmodel,butalsoconducts thestudyof influential factorsaffecting the load ofEVchargingstationsso thatascientificandreasonable input indexsystemisproduced. 2.Methodology 2.1. LionAlgorithmImprovedbyNiche Immune (NILA) 2.1.1. LionAlgorithm(LA) Lionalgorithmisasocialbehavior-basedbionicalgorithmdevelopedbyB.R.Rajakumar in2012. The iterationandgenerationofoptimalsolutionscanberealizedthroughterritorial lion’sbreeding, anditsdefense toothernomadic lions. In thisapproach,everysinglesolutioncorresponds to“Lion”. LAproceeds throughfourmainsteps: population initialization,matingandmutation, territorial defense,andterritorial takeover. Theobjective function issetasEquation(1): minf(x1,x2, · · · ,xn), (n≥1) (1) Step1: Population initialization 356