Page - 402 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies 2018,11, 242 inwhichαl andα∗l are theLangrangemultipliersandcanbedeterminedbysolvingthe followingdual optimizationproblem[46]:⎧⎪⎪⎪⎪⎨⎪⎪⎪⎪⎩ max α,α∗ −ε N ∑ l=1 (α∗l +αl)+ N ∑ l=1 (α∗l −αl)y(l)− 1 2 N ∑ l,t=1 (α∗l −αl)(α∗t−αt)ϕT(x(l))ϕ(x(t)), N ∑ l=1 α∗l = N ∑ l=1 αl, 0<αl,α ∗ l <C, (30) whereC is theregularizationparameterand ε is theerror toleranceparameter. 4.2.AppliedDataSets andExperimentalSetting In this subsection,firstofall, thebuildingenergyconsumptiondatasetswillbedescribed.Next, threedesignfactors thatareutilizedtodeterminetheoptimalstructureof theMDBNwillbeshown. Finally,five indiceswillbegiventoevaluate theperformancesof thepredictivemodels. 4.2.1.AppliedDataSets Twokindsofbuildingenergyconsumptiondatasetsweredownloadedfrom[47]. Thefirstdata set includes34,848samples from2January2010to30December2010.Thedata in thisdatasetwere collectedevery15min inoneretail store inFremont,CA,USA. Wethenaggregatedthemtogenerate thehourlyenergyconsumptiondata. Theseconddatasetcontains22,344samples from4April2009 to 21October2011. Thedata in this setwerecollectedevery60min inoneofficebuilding inFremont, CA,USA.Partsof thesamplesof the twodatasetsaredepicted inFigure7. D E 7LPH RI SDUWV RI WKH VDPSOLQJ GD\V RQH XQLW PLQXWHV 7LPH RI SDUWV RI WKH VDPSOLQJ GD\V RQH XQLW PLQXWHV Figure7.Partsof thesamplesof twodatasets: (a) thefirst500datapointsof theretail store; (b) thefirst 500datapointsof theofficebuilding. 402