Seite - 217 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1900 andxj is themeanvalueandσj thevarianceof the j input factor σj= [ ∑Ni=1 ( xij−xj )2 N−1 ]1/2 . (17) TheStandardizedRegressionCoefficient [20] for the input factor j isdefinedas SRCJ= βjσj σy . (18) Under thepremise that the inputvariablesare independent, theSRCsshowthe importanceof each factor throughmovingeach factor from its expectedvaluebyafixed fractionof its standard deviationwhilekeepingallother factorsat theirexpectedvalues [20]. CalculatingtheSRCsmeans to performtheregressionanalysis,with inputandoutputparametersnormalizedtozeroandstandard deviationone.Apositivesign indicates that the input ispositivelycorrelatedwith theoutput,whilea negativesign indicatesanegativecorrelation. The importanceof these factorscanberankedaccording to theabsolutevalueof theSRCs. 3.CaseStudyandResults 3.1. TheFrameworkof theCaseStudy Theframeworkof thecasestudyisshowninFigure2. Thereal-timemeteorologicalparameters thatwecollectedwere input intotheDesignBuilder (DB),andthesimulatedcoolingloadwasregarded as thereal loadof theofficebuilding.Next, theactualmeteorologicaldataandsimulated loaddata are used as training samples to build a load forecastingmodel based on SVM.Weuse the actual weather data in July as test samples to perform load forecasting to obtain the predicted load P1. Then, theweather forecastdatabeforeandafterprocessingare input into theSVMmodel toobtain predicted loads P2 and P3, which are used for comparing the prediction accuracy of P2 and P3. Sensitivityanalysiswasusedtostudythe factors thathaveasignificant impacton loadforecasting in the inputparameters. The predicted load P1 Real load The predicted load P2 The predicted load P3 The SVM model The DB model Actual weather data Forecast weather data Forecast weather data revised by MCM Actual weather data (from Jun.16 to Sep.15) Figure2.Theframeworkof thecasestudy. In order to verify thevalidity of theMCMand the SVMmodel established, three evaluation indexesareusedtocomparethepredictionresultsbetweenP1,P2andP3,whichincludethefollowing: theMeanAbsolutePercentageError (MAPE)[33], theMeanAbsoluteError (MAE)[34], theRootMean SquareError (RMSE) [34]. Theirdefinitionsandthecalculationresultscanthenbeshownas follows. MAPE= 1 N∑ N i=1 ∣∣∣∣yˆi−yiyi ∣∣∣∣ (19) 217