Seite - 267 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article Short-TermForecastingforEnergyConsumption throughStackingHeterogeneousEnsemble LearningModel MerganiA.Khairalla 1,2,* ID ,XuNing1,NashatT.AL-Jallad1 andMusaabO.El-Faroug3 ID 1 SchoolofComputerScienceandTechnology,WuhanUniversityofTechnology,Wuhan430070,China; xuning@whut.edu.cn(X.N.); jallad@whut.edu.cn(N.T.A.-J.) 2 SchoolofScienceandTechnology,NileValleyUniversity,Atbara346,Sudan 3 FacultyofEngineering,ElimamElmahdiUniversity,Kosti11588,Sudan;musaabgaffar@mahdi.edu.sd * Correspondence:mergani@whut.edu.cn;Tel.:+86-188-7609-0760 Received: 20May2018;Accepted: 13 June2018;Published: 19 June2018 Abstract: Inthereal-life, time-seriesdatacompriseacomplicatedpattern,henceitmaybechallenging to increasepredictionaccuracyratesbyusingmachine learningandconventional statisticalmethods as single learners. This researchoutlines and investigates theStackingMulti-LearningEnsemble (SMLE)model for timeseriespredictionproblemovervarioushorizonswitha focusonthe forecasts accuracy, directionshit-rate, and theaveragegrowth rateof total oil demand. This investigation presentsaflexibleensemble frameworkin lightofblendheterogeneousmodels fordemonstrating andforecastingnonlinear timeseries. TheproposedSMLEmodelcombinessupportvectorregression (SVR),backpropagationneuralnetwork(BPNN),andlinearregression(LR) learners, theensemble architectureconsistsof fourphases: generation,pruning, integration,andensemblepredictiontask. Wehaveconductedanempirical study toevaluateandcompare theperformanceofSMLEusing GlobalOilConsumption (GOC). Thus, the assessment of theproposedmodelwas conducted at singleandmultistephorizonpredictionusinguniquebenchmarktechniques. Thefinal results reveal that theproposedSMLEmodeloutperformsall theotherbenchmarkmethods listed in this studyat various levelssuchaserrorrate, similarity,anddirectionalaccuracyby0.74%,0.020%,and91.24%, respectively. Therefore, thisstudydemonstrates that theensemblemodel isanextremelyencouraging methodologyforcomplex timeseries forecasting. Keywords: timeseriesforecasting;ensemblelearning;heterogeneousmodels;SMLE;oilconsumption 1. Introduction InMachineLearning(ML),ensemblemethodscombinevarious learners tocalculateprediction basedonconstituent learningalgorithms[1]. ThestandardEnsembleLearning(EL)methods include bootstrap aggregating (or bagging) and boosting. RandomForest (RF) [2]; for instance, bagging combines random decision trees and can be used for classification, regression, and other tasks. The effectiveness of RF for regression has been investigated and analyzed in [3]. The boosting method, which builds an ensemble by adding new instances to emphasize misclassified cases, yieldscompetitiveperformancefor timeseries forecasting[4].As themostgenerallyutilizedusage of boosting, Ada-Boost [5] has been comparedwith otherMLalgorithms such as support vector machines (SVM) [6] and furthermore combinedwith this algorithm to additionally enhance the forecastingperformance[7].Also, stacking[8] isan instanceofELmultiplealgorithms. It combines the yieldwhich is produced by various base learners in the first level. In addition, by utilizing ameta-learner, it tries tocombinetheoutcomesfromthesebaselearners inanidealmethodtoaugment thegeneralizationability [9]. Energies2018,11, 1605;doi:10.3390/en11061605 www.mdpi.com/journal/energies267