Seite - 396 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies 2018,11, 242 It isobvious that theextractionof theenergy-consumingpattern, thegenerationof theresidual dataandtheconstructionof theMDBNmodelarecrucial inorder tobuild theproposedhybridmodel. Consequently,wewill introduce themindetail in the followingsubsections. 7UDLQLQJ'DWD 3DWWHUQ ([WUDFWLRQ 3HULRGLFLW\ .QRZOHGJH 0'%1 0RGHO¦ 5HVLGXDO 'DWD +\EULG 0RGHO )LQDO 2XWSXW Figure3.Thestructureof thehybridmodel. 3.2. Extractionof theEnergy-ConsumingPatternsandGenerationof theResidualData Obviously,variousregularpatternsofenergyconsumption(e.g.,daily-periodicity,weekly-periodicity, monthly-periodicityandevenyearly-periodicity) exist indifferent kindsof buildings. In this study, wewill take thedaily-periodicandtheweekly-periodicenergy-consumingpatternsasexamples to introduce themethodforextractingthemfromtheoriginaldata. 3.2.1. TheDaily-PeriodicPattern Fordaily-periodicenergy-consumingpattern, it canbeextractedfromtheoriginal timeseriesby the followingequation: Y¯Ave= [ 1 M M ∑ z=1 yz(1), 1 M M ∑ z=1 yz(2), . . . , 1 M M ∑ z=1 yz(T) ] . (6) Then, theresidual timeseriesYResof thedatasetafter removingthedaily-periodicpatterncanbe generatedas YRes= { Y1−Y¯Ave,Y2−Y¯Ave, · · · ,YM−Y¯Ave } . (7) 3.2.2. TheWeekly-PeriodicPattern Being different from the daily-periodic energy-consuming pattern, the weekly-periodic energy-consumingpattern includes twoparts,whichare thepatternsofweekdaysandweekends. Theweekdaypatternandtheweekendpatterncanberespectivelycomputedas Y¯Weekday= [ 1 M1 M1 ∑ z=1 pz(1), 1 M1 M1 ∑ z=1 pz(2), . . . , 1 M1 M1 ∑ z=1 pz(T) ] , (8) Y¯Weekend= [ 1 M2 M2 ∑ z=1 qz(1), 1 M2 M2 ∑ z=1 qz(2), . . . , 1 M2 M2 ∑ z=1 qz(T) ] , (9) where P= { P1=[p1(1), . . . ,p1(T)], . . . ,PM1 =[pM1(1), . . . ,pM1(T)] } , (10) Q= { Q1=[q1(1), . . . ,q1(T)], . . . ,QM2 =[qM2(1), . . . ,qM2(T)] } , (11) 396