Seite - 39 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1009 Figure13.TheenlargementcomparisonofPeaks5and6fromthecomparedmodels forExample2. Figure14.TheenlargementcomparisonofPeaks7and8fromthecomparedmodels forExample2. For comparisonwithother alternativemodels, Table 7demonstrates the forecastingaccuracy indexes for each comparedmodel. Obviously, theproposedSSVRCCSmodel almost achieves the smallest indexvalues in termsof theMAPE(0.46%),RMSE(126.10), andMAE(80.85), respectively. It is superior to theotherfivecomparedmodels.Onceagain, it indicates that theproposedSSVRCCS modelcouldproducemoreaccurate forecastingperformances. Table7.Forecastingaccuracy indexesofcomparedmodels forExample2. ForecastingAccuracyIndexes SARIMA(9,1,10)×(4,1,4) GRNN(œ=0.07) SSVRCCS SSVRCS SVRCCS SVRCS MAPE(%) 5.16 3.19 0.46 0.86 2.30 3.42 RMSE 1233.09 753.97 126.10 262.02 515.10 886.67 MAE 956.14 577.48 80.85 152.02 426.42 631.40 Finally, twostatistical testsarealsoconductedtoensure thesignificantcontribution in termsof forecastingaccuracy improvement for theproposedSSVRCCSmodel. The test resultsare illustrated in Table8 that theproposedSSVRCCSmodelalmost reachessignificance level in termsof forecasting performance thanotheralternativecomparedmodels. 39