Page - 409 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies 2018,11, 242 4.4.1. Energy-ConsumingPatternof theOfficeBuilding Being similar to the retail store experiment, we utilize Equations (8)–(14) to obtain the weekly-periodicenergy-consumingpatternandtheresidual timeseriesof theofficebuilding. Asmentionedpreviously, theweekly-periodicenergy-consumingpatternshould include two parts,whichare theweekdaypatternand theweekendpattern. Theobtainedweekdaypattern is depicted in Figure 11a, while theweekendpattern is shown in Figure 11b. We can observe that theenergyconsumption inweekends isquitedifferent fromthat inweekdays. After removing the energy-consumingpattern, theresidual timeseriesof theofficebuilding isdemonstrated inFigure11c. This residual timeseries isutilizedto train theMDBNinthehybridmodel. D E F 7LPH RI WKH ZRUNLQJ GD\ RQH XQLW PLQXWHV 7LPH RI DOO WKH VDPSOLQJ GD\V RQH XQLW PLQXWHV 7LPH RI WKH ZHHNHQG RQH XQLW PLQXWHV Figure 11. Periodicity knowledge and the residual time series of the office building data set: (a) the energy-consuming pattern of weekdays; (b) the energy-consuming pattern of weekends; (c) theresidual timeseries. 4.4.2.Configurationsof thePredictionModels Similarly, we run 33 = 27 trials to determine the optimal structure of theMDBNmodel for the office building energy consumptionprediction. The experimental results are listed inTable 4. AsshowninTable4, the trail 13obtains thebestperformance.Consequently, theoptimalstructureof theMDBNinthehybridmodel forofficebuildinghas threehiddenlayers,100hiddenunits ineach layerandfour inputvariables. 409