Seite - 226 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1900 accordingtoFigure10andTable5. Furthermore, throughsensitivityanalysis, itwas foundthatamong theselectedweatherparameters showninFigure11, the factor thathad thegreatest impacton the predictionresultswas the1-h-aheadtemperatureT(h–1)at thepredictionmoment. It isworthnoting thatbothof theresultsP2andP3areobtainedbytheSVMmodel, andtheir uncertainties include twoparts: one is theuncertainty fromtheSVMmodel itself, and theother is theuncertainty fromtheweather forecastdata. Thispapermainly focusesonadopting theMCM toprocessweather forecast data andexplore the impact of theuncertaintyofweather forecast on loadforecasting,not the loadforecastingmodel itself. Theresearch just selects theSVMmodelasan example,because it isacceptedbymost researchersduetogoodperformance.With thedevelopment ofartificial intelligencealgorithmtechnology, theoptimalcombinationofvariousalgorithmsisused for loadprediction [35], which indicates that the accuracy of load forecastingmodel itself can be improvedtosomeextent. Thedata-preprocessingmethodbasedonMCMmakes the forecastingresults closer to theactual loadcompared to thosewithoutprocessing,which is suitable fornotonlyofficebuildingsbutalso other types of buildings. Theprecise load forecasting results are conducive to theHVACsystem operation control. Moreover, theMCMmethod is convenient for application. Historicalweather forecastdataandreal-timemeteorologicaldataareobtainedfromreliableweather forecastingagencies. Inaddition, SPSS isused toanalyzeandobtain theprobabilitydistributionsof theerrorsbetween weather forecastdataandreal-timemeteorologicaldata. Therevisedweatherdataareobtainedby MATLABwiththerelevantprogramsaccordingto theprobabilitydistributionof theerrors. Bothof the toolsare free forapplication. WhenusingtheMCMtoprocessweather forecastdata, it isnecessary toanalyze theprobability distributioncharacteristics that theerrorbetweentheweather forecastdataandactualweatherdata obey. Thecurrentwork is limitedby thesourcesofhistoricalweather forecastdata. The larger the historical samplessizewecollect fromtheweather forecastwebsites, themoreaccurate theprobability distributionfunctionof theerrors, andthenthecloser themodifiedweather forecastdata to theactual weatherdata. In the future,under thecondition that themeteorological forecastdatasourcesaremore widelyavailableandreliable, the1-h-aheadloadforecastingmodelcanbeestablishedtopredict the loadcombinedwiththeMCMfordataprocessing. Itseemsthatmorepreciseresultsof loadpredictions willbeobtained.With thecompletionof follow-upwork, softwareof thedata-preprocessingmethod basedonMCMwillbedeveloped. 5.Conclusions Thispaper investigatedthe influenceof theuncertaintyofweather forecastdataonthecooling load forecast. Here, taking the 24-h-ahead SVMmodel as an example, theMCMwas adopted to preprocessmeteorological forecastdata to improvetheaccuracyof loadforecasting. Itwas indicated that theaccuracyof the loadforecastingwith thedataprocessedbytheMCMisbetter thanthatof the loadforecastingusingthemeteorological forecastdatadirectly,which iscloser to thereal load. Amongtheselectedinputparameters, thefactorsthathavethegreatest impactontheloadforecast areT(h–1)>T(h)>L(d–1,h)>T(h–3)>RH(h)>T(h–2)>RH(h–1)>RH(h–2) in turn. Therefore,wemust improve theaccuracyofmodel inputparameters toreduce the influenceofuncertaintyderivingfrom inputparameterson loadforecasting,especially those influential inputparameters. Author Contributions: Conceptualization, J.Z. and Y.D.; Methodology, J.Z. and Y.D.; Software, Y.D. and X.L.; Validation, Y.D. and X.L.; Formal Analysis, Y.D.; Writing—Original Draft Preparation, Y.D. and X.L.; Writing—Review&Editing, J.Z.andY.D. Funding:This researchwasfundedbytheNaturalScienceFoundationofChina,grantnumber [51508380]. Conflictsof Interest:Theauthorsdeclarenoconflictof interest. 226