Page - 268 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1605 Althoughmultisteppredictionsaredesiredinvariousapplications, theyaremoredifficult tasksthan theone-step,duetolackof informationandaccumulationoferrors. Insomeuniversal forecastingrivalries held lately, different forecastingmethodswereproposed to solve somegenuine issues. In numerous studies,authorscomparedtheperformanceofhybridmodelonlong-termforecasting, for instance, in[10], comparisonresultsdemonstratedthatanensembleofneuralnetworks,suchasmultilayerperceptron(MLP), performedwell in thesecompetitions [10]. Also,Ardakanietal. [11]proposedoptimalartificialneural networks(ANN)modelsbasedonimprovedparticleswarmoptimizationforlong-termelectricalenergy consumption.Regardingthesameaspect thisstudy, [12] introducedamodelnamedthehybrid-connected complexneuralnetwork(HCNN),whichisabletocapturethedynamicsembeddedinchaotic timeseries andtopredict longhorizonsofsuchseries. In[13], researcherscombinedmodelswithself-organizingmaps for long-termforecastingofchaotic timeseries. On the other hand, in short-term forecastingmodels, such asANNandSVM,provide excellent performance for one-step forecasting task [14,15]. However, thesemodels perform poorly or suffer severedegradationwhenappliedtothegeneralmultistepproblems.Aswell, the long-termforecasting modelsaredesignedfor longtimepredictiontasks(for instancemonthlyorweeklytimeseriesprediction). Thatmeanstheymayperformbetter inmultistepforecasting,whileworseinone-stepaheadthanother methods. In general, the performance of combined forecastingmodels (e.g., mixing short-term and long-termapproaches) isbetterwhencomparedtosinglemodels[16]. Therefore,aforecastingcombination canbebenefit fromperformanceadvantagesofshort-termandlong-termmodels,whileavoidingtheir disadvantages. Furthermore,majorstaticcombinationapproaches[17–19]dependonassignafixedweight foreachmodelsuchas(average, inversemean),whiledynamiccombinationsmethodssuchasbaggingand boostinginvestigatedtocombinetheresultsofcomplementaryanddiversemodelsgeneratedbyactively perturbing,reweighting,andresamplingtrainingdata[20,21].Thereforehorizondependentweightsused toavoidtheshortcomingofastaticanddynamiccombinationforshort-andlong-termforecasts[14]. OilConsumption(OC)isasignificantfactorforeconomicdevelopment,whiletheaccuracyofdemand forecasts isanessential factor leadingtotheaccomplishmentofproficiencyarranging.Duetothisreason, energyanalystsareconcernedwithhowtopickthemostsuitableforecastingmethodstoprovideaccurate forecastsofOCtrends [22]. However,numerous techniquescontribute toestimatingtheoildemandin future.Thefieldofenergyproduction,consumption,andpriceforecastinghavebeengainingsignificance asacurrentresearchthemeintheentireenergysectors. For instance,numerousstudies investigatedfoe electricitypriceforecastingsuchasRafał [23], thisreviewarticleaimstoexplainandpartitiontheprimary methodsofelectricitypriceforecasting.Furthermore,Silvanoetal. [24]analyzedelectricityspot-pricesofthe Italianpowerexchangebycomparingtraditionalmethodsandcomputational intelligencetechniquesNN andSVMmodels.Also,NimaandFarshid[25]proposedahybridmethodforshort-aheadpriceforecasting composedofNNandevolutionaryalgorithms. Several studies discussed the issue of time series prediction using different methodologies including statistical methods, single machine learning models, soft computing on ensemble, andhybridmodeling. Statisticalmethodshavebeeninvestigatedfortimeseriespredictionintheenergyconsumptionarea, suchasmovingaverage[26],exponentialsmoothing[27,28],autoregressivemovingaverage(ARMA)[29], andautoregressiveintegratedmovingaverage(ARIMA)models[30]. For instance, theARIMAmodelhas beenintroducedfornaturalgaspriceforecasting[31].However, thesestatistical techniquesdonotyield convincingresults for complicateddatapatterns [32,33]. In this context, theGrayModel (GM) forecast accuracywasenhancedbyusingaMarkov-chainmodel.Theoutcomeofthisstudydemonstratedthatthe hybridGM-Markov-chainmodelwasmoreaccurateandhadahigherforecastaccuracythanGM(1,1) [34]. In fact,neuralnetworksofferapromisingtool forsinglemachine learningmodel in timeseries analysisdueto theiruniquefeatures [35]. Tofurther improvethegeneralizationperformance,ANN modelswere investigated for forecasting futureOC[36]. Another studyexperimentedwithANN models topredict the long-termenergyconsumption[37]. For thesamepurpose,anANNmodelwas appliedto forecast loaddemands in future [38]. 268