Page - 211 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article UncertaintyAnalysisofWeatherForecastDatafor CoolingLoadForecastingBasedontheMonte CarloMethod JingZhao* ID ,YaoqiDuan ID andXiaojuanLiu TianjinKeyLabof IndoorAirEnvironmentalQualityControl,SchoolofEnvironmentalScienceand Engineering,TianjinUniversity,Tianjin300350,China;mgzcdyq@163.com(Y.D.); lxj15869159276@163.com(X.L.) * Correspondence: zhaojing@tju.edu.cn;Tel.:+86-22-87402072 Received: 29 June2018;Accepted: 17 July2018;Published: 20 July2018 Abstract:Recently, thecooling loadforecastingfor theshort-termhasreceivedincreasingattention in thefieldofheating,ventilationandairconditioning(HVAC),whichisconducivetotheHVACsystem operation control. The load forecasting basedonweather forecast data is an effective approach. Themeteorological parameters areusedas thekey inputs of thepredictionmodel, ofwhich the accuracy has a great influence on the prediction loads. Obviously, there are errors between the weather forecast data and the actualweather data, butmost of the existing studies ignored this issue. Inorder todealwith theuncertaintyofweather forecastdatascientifically, thispaperproposes aneffectiveapproachbasedontheMonteCarloMethod(MCM)toprocessweather forecastdata byusing the24-h-aheadSupportVectorMachine (SVM)model for loadpredictionasanexample. Thedata-preprocessingmethodbasedonMCMmakes the forecasting results closer to theactual loadthanthosewithoutprocess,whichreduces theMeanAbsolutePercentageError (MAPE)of load predictionfrom11.54%to10.92%. Furthermore, throughsensitivityanalysis, itwasfoundthatamong theselectedweatherparameters, the factor thathadthegreatest impactonthepredictionresultswas the1-h-aheadtemperatureT(h–1)at thepredictionmoment. Keywords: uncertainty analysis; load forecasting; theMonteCarloMethod (MCM); the Support VectorMachine (SVM)model 1. Introduction Inrecentyears,heating,ventilationandairconditioning(HVAC)systemshavebecomeimportant elements inofficebuildingsandareresponsible foraround40%of theenergyuse inofficebuildings, whichmeansagreatenergy-savingpotential [1].However, theoperationmanagement levelofHVAC systems isgenerally low, and the refrigerationcapacityof theequipmentdoesnotmatchwith the actual demand, resulting in a large energy consumption. Precise load forecasting is the basis of theoptimizationofHVACsystemoperation,which isconducive to formulateanoperationstrategy according to the loadchangeandcan lay the theoretical foundation forenhancing the thermalcomfort andreducingtheenergyconsumptionofofficebuildings.Amongtheinfluential factors,meteorological parametersplayavery important role in thedynamiccooling load,whichhasagreat influenceonthe actualenergyconsumptionofabuilding. In therelevant literatureonbuilding loadforecasting,variouspredictionmodelsareproposedfor loadforecastingandrelatedresearch.XiaandXiangetal. [2]proposedapredictionmodelbasedona radialbasis function(RBF)neuralnetworkto forecastadaily load,whichmainly tooksomeweather parameters intoconsiderationincludingtemperature,humidity,windspeed,atmosphericpressureand soon.Theforecastingresults illustratedthatthemodelhasbetterperformancecomparedwiththeBack Propagation(BP)network.Ruzicetal. [3]put forwardaregression-basedadaptiveweather-sensitive Energies2018,11, 1900;doi:10.3390/en11071900 www.mdpi.com/journal/energies211