Page - 2 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2226 been continuously applied to load forecasting. Furthermore, thehybridizationor combinationof the intelligentalgorithmsalsoprovidesnewmodels to improve the loadforecastingperformances. These hybrid or combinedmodels either employ a novel intelligent algorithm or framework to improve theembeddeddrawbacksorapply theadvantagesof twoof theabovemodels toachieve more satisfactory results. Themodels apply awide rangeof load forecasting approaches andare mainlydividedinto twocategories, traditional forecastingmodelsandintelligent forecastingmodels. 1.2. RelevantLiteratureReviews Conventional load forecastingmodels include exponential smoothingmodels [1], time series models [2], andregressionanalysismodels [3]. Anexponential smoothingmodel is a curvefitting methodthatdefinesdifferentcoefficients for thehistorical loaddata. It canbeunderstoodthataseries with the forecasted loadtimehasa large influenceonthe future load,whileaserieswith the longtime fromtheforecasted loadhasasmall influenceonthe future load[1]. The timeseriesmodel isapplied to loadforecasting,which ischaracterizedbyafast forecastingspeedandcanreflect thecontinuity of load forecasting, but requires the stability of the time series. Thedisadvantage is that it cannot reflect the impact of external environmental factors on load forecasting [2]. The regressionmodel seeksacausal relationshipbetweenthe independentvariableandthedependentvariablesaccording to thehistorical loadchange law,determining the regressionequation, and themodelparameters. Thedisadvantageof thismodel is that therearetoomanyfactorsaffectingtheforecastingaccuracy. It is notonlyaffectedbytheparametersof themodel itself,butalsobythequalityof thedata.Whenthe external influence factors are toomanyor the relevant influent factordata aredifficult to analyze, theregressionforecastingmodelwill result inhugeerrors [3]. Intelligent forecasting models include the wavelet analysis method [4,5], grey forecasting theory [6,7], the neural networkmodel [8,9], and the support vector regression (SVR)model [10]. In load forecasting, thewavelet analysismethod is combinedwith external factors to establish a suitable loadforecastingmodelbydecomposingthe loaddata intosequencesondifferentscales [4,5]. The advantages of the greymodel are easy to implement and there are fewer influencing factors employed. However, the disadvantage is that the processed data sequence hasmore grayscale, whichresults in large forecastingerror [6,7]. Therefore,whenthismodel isapplied to loadforecasting, onlya fewrecentdatapointswouldbeaccurately forecasted;moredistantdatacouldonlybereflected as trendvaluesandplannedvalues [7].Dueto thesuperiornonlinearperformances,manymodels based on artificial neural networks (ANNs) have been applied to improve the load forecasting accuracy[8,9]. Toachievemoreaccurateforecastingperformance, thesemodelsandothernewornovel forecastingapproacheshavebeenhybridizedorcombined[9]. Forexample,anadaptivenetwork-based fuzzy inferencesystemiscombinedwithanRBFneuralnetwork[11], theMonteCarloalgorithmis combinedwith theBayesianneuralnetwork[12], fuzzybehavior ishybridizedwithaneuralnetwork (WFNN) [13], a knowledge-based feedback tuning fuzzy system is hybridizedwith amulti-layer perceptronartificialneuralnetwork(MLPANN)[14], andsoon.However, theseANNs-basedmodels sufferfromsomeseriousproblems,suchastrappingintolocaloptimumeasily, itbeingtime-consuming toachievea functionalapproximation,andthedifficultyof selecting thestructuralparametersofa network[15,16],which limits itsapplication in loadforecastingtoa largeextent. The SVRmodel is based on statistical learning theory, as proposed byVapnik [17]. It has a solidmathematical foundation, a better generalization ability, a relatively faster convergence rate, and canfindglobal optimal solutions [18]. Because the basic theory of the SVRmodel is perfect andthemodel isalsoeasytoestablish, ithasattractedextensiveattentionfromscholars in the load forecasting fields. In recent years, some scholars have applied the SVRmodel to the research of loadforecasting[18]andachievedsuperior results.Onestudy[19]proposes theEMD-PSO-GA-SVR model to improvethe forecastingaccuracy,byhybridizingtheempiricalmodedecomposition(EMD) with twoparticleswarmoptimization(PSO)andthegeneticalgorithm(GA). Inaddition,amodified versionof theSVRmodel,namelytheLS-SVRmodel,onlyconsidersequalityconstraints insteadof 2