Seite - 269 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1605 However, ANNsyieldmixed resultswhendealingwith linear patterns [39], it is difficult to obtainhighaccuracyratesofpredictorsbyusing thesinglemethod,eitherstatisticalorMLtechniques individually. In order to avoid the limitations associatedwith the individualmodels; researchers suggestedahybridmodelwhich combines linear andnonlinearmethods toyieldhighprediction accuracyrates [32,39]. Several studies investigatedhybridmodeling tooptimize theparametersof the ANN[40].Hence the improvedperformanceofartificialbeecolony(ABC-LM)overotheralternatives hasbeendemonstratedonbothbenchmarkdataandOCtimeseries. Similarly,anNN,combinedwiththreealgorithmsinahybridmodel,thenoptimizedbyusingagenetic algorithmwasusedtoestimateOC;theoutcomedemonstratedtheefficiencyofthehybridmodeloverall benchmarkmodels[41].Moreover,aresearcher in[42]proposedageneticalgorithm—grayneuralnetwork (GA-GNNM)hybridmodel toavoid theproblemofover-fitting,byexamininghybridversusa totalof 26combinationmodels.Authorsconcludedthatthehybridmodelsprovideddesirableforecastingresults, comparedtotheconventionalmodels. Also, theGAhasmoreflexibility inadaptingNNparameters to overcometheperformanceinstabilityofneuralnetworks[22]. In thesamecontext,hybridmodelswere investigatedtosolveprediction intervalsanddensities problems,andhavebecomemorecommon.AsshowninHansen[43] fuzzymodelcombinedwith neuralmodels, this combination increased the computation speed, and the coverage is extended. Thus, theproblemof the narrowprediction intervals is resolved. Similarly, in [44] theprediction intervalalsoconcernedwithblendofneuralnetworksandfuzzymodels todetermine theoptimal orderforthefuzzypredictionmodelandestimateitsparameterswithgreateraccuracy. Sinceprediction intervalsandforecastdensitieshavebecomemorepopular,manytypesof researchhavebeendone about how todetermine the appropriate input lag, for this purpose, the fuzzy time seriesmodel suggested increasingaccuracybysolvingtheproblemsofdatasize (sampling)andthenormality [45]. Regardingthesameaspect,EfendiandDerisextendedanewadjustmentof the interval-lengthand thepartitionnumberof thedataset, this studydiscussedthe impactof theproposedinterval length inreducingthe forecastingerrorsignificantly,aswellas themaindifferencesbetweenthe fuzzyand probabilisticmodels [46]. Finally,asaconclusionfromtheabovestudies,hybridmethodsgiveoffan impressionofbeing an astoundingway to combinepredictions of several learning algorithms. Thehybrid regression modelsgivepreferredpredictiveaccuracyoveranysinglelearner.Nonetheless, therewasnodistinctive waytomerge theoutcomeforecastsof individualmodels. In thispaper, thegoal is to introduceanovelELframeworkthatcanreducemodeluncertainty, enhancemodel robustness,andenhance forecastingaccuracyonoildatasets, improvemodelaccuracy, beingdefinedashavinga lowermeasureof forecasting error. Themost importantmotivation for combiningdifferent learningalgorithmsisbasedupontheassumptionthatdiversealgorithmsusing differentdatarepresentations,dissimilarperceptions,andmodellingmethodsareexpectedtoarrive atoutcomeswithdifferentprototypesofgeneralization[47]. Inaddition, todate, comparatively few researcheshaveaddressedensembles fordifferent regressionalgorithms[48]. Wedemonstrate that theOCframeworkcansignificantlyoutperformthecurrentmethodologies ofutilizingthesingleandclassicensemble forecastingmodels insingleandmultistepperformance. Although the idea is straightforward, it is yet a robust approach, as it canoutperformtheaverage model, asonedoesnotknowaprioriwhichmodelwillperformbest. Themeritsof thisproposed methodology are analyzed empirically by first describing the exact study design and after that, assessing theperformanceofvariousensemblesofdifferentOCmodelsontheGOC.Theseoutcomes are thencomparedto theclassicalapproach inthe literature,whichtakes thecalibratedmodelwith the lowestmeasureof forecastingerroronthecalibrationdatasetat thehorizon(1-ahead) toOCof the samedatasetat thehorizon t=n (10-ahead). In summary, the developed ensemble model takes full advantage of each component and eventually achieves final success in energy consumption forecasting. Themajor contributions of thispapercometherefore fromthreedimensionsas follows: 269