Seite - 213 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1900 whichhaveagreat impactontheactualenergyconsumptionof thebuilding. It iseffective touse the weather forecastdata topredict thebuilding load inadvanceandadjust theair conditioningunits in timeaccording to the forecast loads for thepurposesof improvementof the indoorcomfortand reductionof building energy consumption. If theuncertainty ofweather forecast data is ignored, itmaycauseerrors inmodel inputs,whichreduces theaccuracyof the forecast load. Thepaperfillsa gapin termsof thecorrectionof theuncertaintyofweather forecastdata. This paper explored the impact ofweather forecast uncertainty on load forecasting, and the Monte Carlo Method (MCM) was used to modify the input parameters of the model for load forecasting,whichcan increase theaccuracyof the load forecastingbeforeandafter thecorrection. Furthermore, thesensitivityanalysiswasadoptedtoexplore the factors thathaveagreat impacton loadforecastingresults. Thecontentsof thepaperareas follows. Section2presentsageneraloverviewof theprinciplesof theMCM,theSVMandsensitivityanalysis. Section3presentsacasestudy, inwhichthiscasestudy isused to illustratehowthemethodologycanbeapplied tostudy the impactofuncertaintyof the weather forecastdataon loadprediction,andthemainfactorscontributing to the loadpredictionare identifiedthroughasensitivityanalysis. Section4presentsadiscussionof theresults, aswellassome proposals for futureresearch. Section5summarizes twoimportantconclusions in theresearch. 2.Methodology In thispaper, theMCMisusedtoanalyze theuncertaintiesofweather forecastingparameters, and the model based on SVM is established to forecast the cooling load of an office building. Inaddition, theStandardizedRegressionCoefficient(SRC)methodforsensitivityanalysisis introduced comprehensively. Theflowchart showninFigure1depicts themainsteps indevelopingtheresearch, whichfacilitates theunderstandingof theproposedapproach. The improvement of load prediction accuracy Sensitivity analysis by the SRCs method The predicted load P3 The predicted load P1 The predicted load P2 Weather forecast data revised by MCM Weather forecast data SVM model The model built by DesignBuilder Actual weather data Real load Figure1.Theframeworkof theresearchmethods. 2.1. TheMCMofRandomSampling inProcessingWeatherForecastData TheMCM, also called a statistical simulationmethod, is an important numerical calculation methodguidedbyprobabilitystatistics theorydueto thedevelopmentofscience technologyandthe inventionof electronic computers in themid–1940s. It is aneffectiveway touse randomnumbers to solvemany problems. TheMonte Carlo simulation is amethod of studying the distribution characteristics by setting up a stochastic process and calculating the estimates and statistics of parameters. Specifically, the reliability of the system is too complex, and it is difficult to establish anaccuratemathematicalmodel forreliabilityprediction.Whenthemodel is inconvenient toapply, 213