Page - 33 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1009 3.2.4. ForecastingResultsandAnalysis forExample1 To compare the improved forecasting performance of the tent chaotic mapping function, aSVRwith theoriginalCSalgorithm(without the tentchaoticmappingfunction),namely theSVRCS model, will also be taken into comparison. Therefore, according to the rolling-based procedure mentionedabove,byusingthe trainingdataset fromExample1 (mentionedinSection3.1) toconduct the trainingwork, and theparameters forSVRCSandSVRCCSmodelsareeventuallydetermined. These trainedmodelsare furtherusedtoforecast theelectric load. Then, the forecastingresultsandthe suitableparametersofSVRCSandSVRCCSmodelsare listed inTable1. It is clearly indicatedthat the proposedSVRCCSmodelhasachievedsmaller forecastingperformances in termsof the forecasting accuracy indexes,MAPE,RMSE,andMAE. Table 1. Three parameters of SVRCS and SVRwith chaotic cuckoo search (SVRCCS)models for Example1. EvolutionaryAlgorithms Parameters MAPEofTesting(%) RMSEofTesting MAEofTesting σ C ε SVRCS 1.4744 17,877.54 0.3231 2.63 217.19 151.72 SVRCCS 0.5254 5,885.65 0.7358 1.51 126.92 87.94 AsshowninFigure3, theemployedelectric loaddatademonstratesseasonal/cyclic changing tendency inExample1. Inaddition, thedata recording frequency isonahalf-hourbasis, therefore, to comprehensively reveal the electric load changing tendency, the seasonal length is set as 48. Therefore, thereare48seasonal indexes for theproposedSVRCCSandSVRCSmodels. Theseasonal indexes for eachhalf-hour are computedbasedon the 576 forecastingvalues of the SVRCCSand SVRCSmodels inthetraining(432forecastingvalues)andvalidation(144forecastingvalues)processes. The48seasonal indexes for theSVRCCSandSVRCSmodelsare listed inTable2, respectively. Table2.The48seasonal indexes forSVRCCSandSVRCSmodels forExample1. Time Points Seasonal Index(SI) Time Points Seasonal Index(SI) Time Points Seasonal Index(SI) Time Points Seasonal Index(SI) SVRCCS SVRCS SVRCCS SVRCS SVRCCS SVRCS SVRCCS SVRCS 00:00 0.9615 0.9201 06:00 1.0360 1.0536 12:00 1.0025 1.0076 18:00 1.0071 1.0176 00:30 0.9881 0.9241 06:30 1.0518 1.0729 12:30 0.9960 1.0032 18:30 1.0034 1.0109 01:00 0.9893 0.9401 07:00 1.0671 1.0924 13:00 0.9935 0.9992 19:00 0.9694 0.9767 01:30 0.9922 0.9729 07:30 1.0394 1.0810 13:30 0.9975 1.0022 19:30 0.9913 0.9875 02:00 0.9919 0.9955 08:00 1.0088 1.0575 14:00 1.0026 1.0083 20:00 0.9820 0.9812 02:30 0.9948 0.9980 08:30 1.0076 1.0322 14:30 1.0015 1.0088 20:30 0.9789 0.9700 03:00 0.9950 0.9998 09:00 1.0004 1.0148 15:00 1.0000 1.0070 21:00 0.9830 0.9641 03:30 0.9915 0.9961 09:30 0.9903 0.9982 15:30 1.0022 1.0089 21:30 0.9780 0.9547 04:00 1.0082 1.0129 10:00 1.0031 1.0067 16:00 1.0033 1.0115 22:00 0.9906 0.9622 04:30 1.0075 1.0176 10:30 0.9912 0.9981 16:30 1.0097 1.0173 22:30 0.9932 0.9778 05:00 1.0124 1.0245 11:00 0.9928 0.9973 17:00 1.0098 1.0188 23:00 0.9659 0.9645 05:30 1.0139 1.0253 11:30 0.9967 1.0025 17:30 1.0053 1.0164 23:00 0.9601 0.9348 The forecasting comparison curves of six models, including the SARIMA(9,1,8)×(4,1,4), GRNN(σ = 0.04),SSVRCCS,SSVRCS,SVRCCS,andSVRCSmodelsmentionedaboveandactual valuesareshowninFigure4. It illustrates that theproposedSSVRCCSmodel iscloser to theactual electric loadvaluesthanothercomparedmodels. Tofurther illustratethetendencycapturingcapability oftheproposedSSVRCCSmodelduringtheelectricpeakloads,Figures5–8areenlargementsfromfour peaks inFigure4 toclearlydemonstratehowcloser theSSVRCCSmodelmatches to theactualelectric loadvalues thanotheralternativemodels. Forexample, foreachpeak, the redreal line (SSVRCCS model)always followscloselywith theblackreal line (actualelectric load),whetherclimbingupthe peakorclimbingdownthehill. 33