Page - 212 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1900 short-termload-forecastingalgorithm.Thisalgorithmwasusedfor the loadpredictionof theElectric PowerUtility of Serbia. Wi [4] presented a fuzzypolynomial regressionmethod for holiday load prediction combinedwith the dominantweather feature, and pointed out that it was pivotal to select thepreviousdata relevant to the givenholiday for improving the accuracyof holiday load forecasting. Support VectorMachines (SVMs) have beenwidely applied in the field of pattern recognition,bioinformatics,andotherartificial intelligencerelevantareas tosolvetheclassificationand regression issues; thesearecalledSupportVectorClassification(SVC)andSupportVectorRegression (SVR).Particularly,alongwithVapnik’s ε-insensitive loss function, theSVMalsohasbeenextended to solvenonlinear regressionestimationproblemsbySVR. It hasbeenwidelyused inmanyfields involvingpredictionproblems, suchasfinancial industry forecasting[5–7], engineeringandsoftware fieldforecasting[8], atmosphericscience forecasting[9]andsoon. Furthermore, theSVRmodelhas alsobeensuccessfullyappliedtopredict thepower load[10]. Theselectionof the threeparameters (C, ε, andσ) in theSVRmodel influences thepredictionaccuracysignificantly. Manystudieshave givenrecommendationsonappropriate settingofSVRparameters [11]. But thosemethodsdonot comprehensivelyconsider the interactioneffectsamongthe threeparameters. Thus, the intelligent algorithmsareadopted todetermineappropriateparametervalues. Barmanet al. [12]proposeda regionalhybridSTLFmodelutilizingSVMwithanewtechniquetoevaluate its suitableparameters andpointedout that theGOA-SVMmodel is targeted for forecasting the loadunder local climatic conditions. Lietal. [13] investigate the feasibilityofusingLeastSquaresSupportvector regression (LS-SVR) to forecastbuildingcooling load. Theevaluationof the tests illustratedthat theSVRmodel with theParticleSwarmOptimization(PSO)hasagoodgeneralizationperformance. Atpresent, the researchon the inputsof thepredictionmodelmainly involves theoptimized selection of input parameters. Duanmuet al. [14] proposed a simplifiedpredictionmodel of the cooling loadbasedonthehourlycooling loadcoefficientmethodandanalyzedthevarious influential factorsof thecooling load. Theypointedout thatoutdoor temperature is thekey influential factorof the cooling load. Wang et al. [15] researched the influence of climate change on the heating and the cooling (H/C) energy requirements of residential houses, which is from cold to hot humid in five regional climates of Australia. They pointed out that the impacts of significant climate change on H/C energy requirements may occur during the lifecycle of existing housing stock. Jiang[16]consideredthat theaccuratepredictionofbuildingthermalperformance isdependenton meteorologicaldatasuchasdry-bulb temperature, relativehumidity,windspeedandsolar radiation toa largeextent. Chenetal. [17] selecteddifferentmeteorologicalvariablesas inputs fordifferent time scales,usingbuildingdynamicssimulationto forecast theenergydemandforcoolingandheatingof residentialbuildings. Petersenetal. [18]analyzedtheimpactofuncertaintyontheindoorenvironment. Indeed,onlya fewstudieshaveformallydealtwith the issueofuncertainty in loadforecasting. Forexample,Sarjiya [19]adoptedadecisionanalysismethodtohandle theuncertaintyof the load forecast inpowersystemsfortheaimofoptimizationoftheoperatingstrategy.Domínguez-Muñoz[20] proposed a new approach based on stochastic simulationmethods to research the impact of the uncertaintyof the internaldisturbanceonthepeakcooling loadinthebuildings.Douglasetal. [21] put forward amethod to analyze the risk of short-termpower systemoperational planningwith theelectrical load forecastuncertainty.MacDonald [22] focusedon theproblemofquantifying the effectofuncertaintyonthepredictionsmadebysimulation tools. Twoapproaches includingexternal andinternalmethodswereusedtoquantify thiseffect.Domínguez-Muñozetal. [23]quantifiedthe uncertainty that canbe expected in the thermal conductivity of insulationmaterials in the lackof specific experimentalmeasurements. Stenet al. [24] analyzed the influenceof theuncertainties of temperaturestratificationandpressurecoefficientsonbuildings in termofnaturalventilation through anexpert reviewprocess. Overviewingthepreviousresearch, fewstudieshavepaidattentiontotheinfluenceofuncertainty ofweather forecast data on the load forecasting. However, external disturbance factors such as meteorological parameters play avery important role in thedynamic cooling loads of a building, 212