Page - 220 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1900 0 10 20 30 40 50 60 0 50 100 150 200 250 300 350 Time (hour) Measured data Simulated data Relative error Figure5.Comparisonbetweenthesimulatedheating loadandthemeasuredheatingsupply. 3.4. TheSVMModel andValidation Thefinal inputparametersof thedynamiccooling loadforecastingmodel for theconstruction of thisproject still need tobedetermined in combinationwith theweather forecast and theactual situation. Forthe24-h-aheadloadforecastingmodel, it isdifficult toobtaininformationabouthistorical loadsandsolar radiationvalues for the1-h to3-hahead, soweselectweatherdataandloaddata from 16 June to15Septemberas trainingsamples (576observationvalues) to establisha supportvector machinemodel. Table2givessomeexamplesof trainingsamples. Table2.Someexamplesof trainingsamples. Time Output Inputs L(h) L(d–1,h) T(h) T(h–1) T(h–2) T(h–3) RH(h) RH(h–1) RH(h–2) 7/3/9:00 −772.61 −584.99 27.4 26.0 25.2 24.7 78.9 84.2 87.8 7/3/10:00 −813.45 −649.59 28.3 27.4 26.0 25.2 75.1 78.9 84.2 7/3/11:00 −861.98 −700.47 28.3 28.3 27.4 26.0 74.3 75.1 78.9 7/3/12:00 −770.72 −660.08 28.5 28.3 28.3 27.4 71.8 74.3 75.1 7/3/13:00 −753.09 −723.44 28.8 28.5 28.3 28.3 71.3 71.8 74.3 7/3/14:00 −881.99 −876.29 29.3 28.8 28.5 28.3 68.2 71.3 71.8 7/3/15:00 −884.55 −844.72 29.2 29.3 28.8 28.5 68.6 68.2 71.3 7/3/16:00 −866.96 −824.41 28.8 29.2 29.3 28.8 69.2 68.6 68.2 7/3/17:00 −829.87 −815.52 28.5 28.8 29.2 29.3 70.2 69.2 68.6 L(d–1,h) represents thehistorical loadat thesamemomentwepredict for thepreviousday. T(h),T(h–1),T(h–2), T(h–3)are thedry-bulb temperatureat themomentweforecastandthe timeof the1–h to3–hahead, respectively, RH(h),RH(h–1),RH(h–2)are therelativehumidityat themomentweforecastandthe1–h to2–hahead, respectively. The particle swarm optimization algorithm is used to optimize the parameters of the support vector machine, where we set ε = 0.1, C [0.1,100], g [0.01,100], where g is γ in Equation (10). The particle swarmoptimization algorithmhyper-parameter optimization results are: BestC = 17.9873, Best g = 0.01, CVmse= 0.0068. Figure6 shows the resultsof thefitness functionof theparticleswarmoptimizationalgorithm. 220