Seite - 412 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies 2018,11, 242 D E F G H 7KH YDOXHV RI WKH UHVLGXDO HUURUV NZK 7KH YDOXHV RI WKH UHVLGXDO HUURUV NZK 7KH YDOXHV RI WKH UHVLGXDO HUURUV NZK 7KH YDOXHV RI WKH UHVLGXDO HUURUV NZK 7KH YDOXHV RI WKH UHVLGXDO HUURUV NZK Figure 13. Prediction error histograms of the five predictors constructed by utilizing the energy-consumingpattern: (a)hybridDBNmodel; (b)BPNN;(c)GRBFNN;(d)ELM;and(e)SVR. Then, in order to examine the superiority of the hybridmodel for the office building energy consumptionprediction, thefivepredictionmodelsarecomparedunder theconsiderationofdifferent data types (theoriginalandresidualdata). Experimental resultsaredemonstrated inTable5. Table5.Theperformancesof thefivemodelswithdifferentdata types for theofficebuildingenergy consumptionprediction. Methods DataType MAE(kwh) MRE(%) RMSE(kwh) r R2 MDBN Residualdata 2.09 11.62 3.54 0.97 0.93Originaldata 2.32 11.50 4.19 0.95 0.90 BPNN Residualdata 2.57 12.64 4.04 0.96 0.93Originaldata 3.85 23.21 4.75 0.95 0.91 GRBFNN Residualdata 2.54 12.62 4.39 0.95 0.91Originaldata 4.35 21.94 5.98 0.93 0.87 ELM Residualdata 3.50 17.18 4.92 0.96 0.92Originaldata 4.61 25.52 5.92 0.90 0.82 SVR Residualdata 3.23 14.89 4.98 0.94 0.88Originaldata 6.13 34.42 7.55 0.92 0.85 412