Seite - 328 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1449 4.CaseStudy Baseonthedaily load,meteorologicaldataandoperationinformationofane-buschargingstation inBaoding,China, in2017,acasestudywascarriedout for thepurposeofdemonstratingtheefficiency of theproposedmodel in loadforecastingfore-buschargingstation. The loaddatawasprovidedby StateGridHebeiElectricPowerCompany inChina, and the inputdatawasprovidedby the local meteorological department. This paper adoptsMatlabR2014b (GamaxLaboratory SolutionsKft., Budapest,Hungary) toprogram, andas for the testplatformenvironment, an IntelCore i5-6300U (Intel Corporation, SantaClara, CA,USA), 4Gmemory andWindows 10 Professional (Microsoft corporation,Redmond,WA,USA)Editionsystemwasused. Inorder toeliminate theparticularityof the targetdaysandexamine thegeneralizationperformanceof theestablishedtechnique, thedata for onedayfromeachof thefourseasonswasselectedas test samples; that is,April15, July15,October15 andJanuary15werechosenas test samples forspring, summer,autumnandwinter, respectively. 4.1. InputSelectionandPre-Processing Basedontheanalysisof loadcharacteristics inthee-buschargingstationinSection2,asetofeight variableswasusedas the input, includingdaytype,maximumtemperature,minimumtemperature, weather condition, the accumulated daily number of charged e-buses and the loads at the same moment in theprevious threedays.Dayscanbedivided into threecategories:workdays (Mondayto Friday),weekends (SaturdayandSunday)andlegalholidayswerevaluedat1,0.5and0, respectively. Weatherconditionswereseparatedinto twotypes,wheresunnyandcloudydayswerevaluedat1, andrainyandsnowydayswerevaluedat0.5. The loadsat thesamemoment in theprevious three daysrefer to thosenearest thepredicteddayinsimilarsamplesafterclusteringaccordingto therule that“Everything lookssmall in thedistanceandisbigonthecontrary.”Thetemperature, loaddata, anddailyaccumulatedchargede-busesshouldbenormalizedaspresented inEquation(1). 4.2.ModelPerformanceEvaluation It’s important to effectively evaluate the load forecasting results for e-bus charging stations, andtheperformanceof thepredictionmodels isusuallyassessedbystatistical criteria: therelative error (RE), rootmean square error (RMSE),mean absolute percentage error (MAPE) andaverage absoluteerror (AAE).Thesmaller thevaluesof these four indicatorsare, thebetter the forecasting performance is. Inaddition, the indicatorsnamedRMSE,MAPEandAAEcanreflect theoverall error of thepredictionmodel and thedegree of error dispersion. The smaller the values of these three indicatorsare, themoreconcentratedthedistributionoferrors is. Thefourgenerallyadoptederror criteriaaredisplayedas follows: (1)Relativeerror (RE) RE= xˆi−xi xi ×100% (23) (2)Rootmeansquareerror (RMSE) RMSE= √ 1 n n ∑ i=1 ( xˆi−xi xi ) 2 (24) (3)Meanabsolutepercentageerror (MAPE) MAPE= 1 n n ∑ i=1 |(xˆi−xi)/xi| ·100% (25) 328