Seite - 270 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1605 1. Inthisstudy,wedevelopanewensembleforecastingmodel thatcanintegrate themeritsofsingle forecastingmodels toachievehigher forecastingaccuracyandstability. Wehave introducedanovel theoretical frameworkhowtopredictOC.Althoughtheensemble concept is more demanding regarding computational requirements, it can significantly outperform the best performing model (SVR) of individual models. While the idea is straightforward, it isyeta robustapproach,as it canoutperformthe linearcombinationmethods, asonedoesnotknowaprioriwhichmodelwillperformbest. 2. Theproposedensemble forecastingmodelaimstoachieveeffectiveperformance inmulti-stepoil consumptionforecasting. Multi-step forecastingcaneffectivelycapture thedynamicbehaviorofoil consumption in the future,which ismorebeneficial toenergysystemsthanone-step forecasting. Thus, this study buildsacombinedforecastingmodel toachieveaccurateresults formulti-stepoil consumption forecasting,whichwillprovidebetterbasic forenergyplanning,productionandmarketing. 3. Thesuperiorityof theproposedensemble forecastingmodel isvalidatedwell inarealenergy consumptiondata. The novel ensemble forecasting displays its superiority compared to the single forecasting model and classic ensemblemodels, and theprediction validity of thedeveloped combined forecastingmodel demonstrates its superiority in oil consumption forecasting compared to classical ensemblemodels (AR,Bagging)andthebenchmarksinglemodels (SVR,BPNNandLR) aswell. Therefore, thenewdevelopedforecastingmodelcanbewidelyused inall temporaldata applicationprediction. 4. Aperceptivediscussion isprovidedinthispaper to furtherverify the forecastingefficiencyof the proposedmodel. Fourdiscussionaspectsareperformed,whichincludethesignificanceof theproposedforecasting model, thecomparisonwithsinglemodels,andclassicalensemblemethods, thesuperiorityof thedeveloped forecastingmodel’s stability,whichbridge theknowledgegap for the relevant studies,andprovidemorevaluableanalysisandinformationforoil consumptionforecasting. The structure of the paper is organized into five sections: Section 2 is devoted to describing proposedmethodsdesign. Section3presentstheexperimentalresults. Section4offerstheconsumption prediction analysis anddiscussion. Section 5describes the conclusion and further suggestion for futureworks. 2.MaterialsandMethods 2.1. ProposedFramework InSection1,reviewedtheliteratureinthreedifferentareas(i.e., single,hybrid,andsoftcomputing onensemble). While thehybridmodeling literature advanced significantly over the last 20years, theresearchonminimizing forecasterror,modeluncertainty, andhybridmethods is still relatively limitedsofar. Tothebestofourknowledge,noattemptsexistyetofcombiningthesedifferentareas, byusingELmethods toreduce the issuesofOCtasks (seeTable1 forasummary). In particular, wewill outline a very general theoretical framework to calibrate and combine heterogeneousMLmodelsusingensemblemethods. Itsmodularity isdisplayedinFigure1andallows forflexible implementationregardingbasemodels, forecasting techniques,andensemblearchitecture. Forapracticalapplicationof thismethod,wehavesplit theStackingMulti-LearningEnsemble (SMLE) frameworkinto fourmainphasesandwilldescribe themincludingtheirsub-steps in furtherdetail as follows. 270