Page - 414 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies 2018,11, 242 Ö \ \ 5 U Ö \ \ 5 U $FWXDO YDOXHV NZK D $FWXDO YDOXHV NZK E Figure14. Scatterplotsof theactual andpredictedvaluesof theenergyconsumptions in the retail building(a) andtheofficebuilding(b). 5.Conclusions In thispaper,ahybridmodel ispresentedto further improve thepredictionaccuracy forbuilding energyconsumptionprediction. Theproposedmodelcombines theMDBNmodelwith theperiodicity knowledge toobtain thefinalpredictionresults. The theoretical contributionsof this studyconsist of twoaspects: (1) theperiodicityknowledgewasextractedandencodedinto thepredictionmodel. In addition, the prediction accuracy can be greatly improved throughutilizing this kind of prior knowledge; (2)anovel learningalgorithmthatcombines thecontrastivedivergencealgorithmand the least squaresmethodwasproposed tooptimize theparameters of theMDBN.This is thefirst time that theDBN is applied to the building energy consumptionprediction.On the other hand, this study applied the proposed approach to the energy consumptionprediction of twokinds of buildings. Experimental andcomparison results verified theeffectiveness andsuperiorities of the proposedhybridmodel. Asiswellknown,manykindsoftimeseriesdata,e.g., thetrafficflowtimeseriesandtheelectricity consumptiontimeseries,havetheperiodicityproperty. Thehybridmodelcanbeexpectedtoyield betterperformance in thepredictionsofsuchtimeseries. In the future,wewill extendourapproach to these applications. On theother aspect, our studyonly focuseson thedata science that tries to utilize thedata to realize theenergy-consumptionpredictionwithout consideringanyscientificor practical informationofenergyrelatedprinciples. Theoretically, theenergyrelatedprinciplesarevery helpful to improvethepredictionperformance.Wearenowexploringthestrategies toconstruct the novelhybridpredictionmodels throughcombining theenergyrelatedprinciplesandobserveddata to further improvethepredictionaccuracy. Acknowledgments: This work is supported by the National Natural Science Foundation of China (61473176, 61105077, 61573225), and theNatural Science Foundation of ShandongProvince forYoungTalents inProvinceUniversities (ZR2015JL021). AuthorContributions:ChengdongLi, JianqiangYiandYishengLvhavecontributedtodeveloping ideasabout energy consumptionprediction and collecting thedata. ZixiangDing andGuiqingZhangprogrammed the algorithmandtested it.Allof theauthorswere involvedinpreparingthemanuscript. Conflictsof Interest:Theauthorsdeclarenoconflictof interest. 414