Page - 119 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article HybridShort-TermLoadForecastingSchemeUsing RandomForestandMultilayerPerceptron† JihoonMoon1,YongsungKim2,MinjaeSon1 andEenjunHwang1,* 1 SchoolofElectricalEngineering,KoreaUniversity,145Anam-ro,Seongbuk-gu,Seoul02841,Korea; johnny89@korea.ac.kr (J.M.); smj5668@korea.ac.kr (M.S.) 2 SoftwarePolicy&ResearchInstitute (SPRi), 22,Daewangpangyo-ro712beon-gil,Bundang-gu,Seongnam-si, Gyeonggi-do13488,Korea;kys1001@spri.kr * Correspondence: ehwang04@korea.ac.kr;Tel.:+82-2-3290-3256 † Thispaper isanextendedversionofourpaperpublishedinProceedingsof the2018IEEEInternational ConferenceonBigDataandSmartComputing(BigComp),Shanghai,China,15–18 January2018. Received: 31October2018;Accepted: 21November2018;Published: 25November2018 Abstract: Astable power supply is very important in themanagement of power infrastructure. Oneof thecritical tasks inaccomplishing this is topredictpowerconsumptionaccurately,which usuallyrequiresconsideringdiverse factors, includingenvironmental, social, andspatial-temporal factors.Dependingonthepredictionscope,buildingtypecanalsobeanimportant factorsince the sametypesofbuildingsshowsimilarpowerconsumptionpatterns. Auniversitycampususually consistsof severalbuilding types, includinga laboratory, administrativeoffice, lecture room,and dormitory.Dependingonthe temporalandexternalconditions, theytendtoshowawidevariation in theelectrical loadpattern. Thispaperproposesahybridshort-termload forecastmodel foran educationalbuildingcomplexbyusingrandomforestandmultilayerperceptron. Toconstruct this model,we collect electrical loaddata of six years fromauniversity campus and split them into training, validation, and test sets. For the training set,we classify thedatausing adecision tree with inputparameters includingdate, dayof theweek, holiday, andacademicyear. In addition, weconsidervariousconfigurations for randomforestandmultilayerperceptronandevaluate their prediction performance using the validation set to determine the optimal configuration. Then, weconstructahybridshort-termloadforecastmodelbycombiningthe twomodelsandpredict the dailyelectrical loadfor the test set. Throughvariousexperiments,weshowthatourhybrid forecast modelperformsbetter thanotherpopularsingle forecastmodels. Keywords: hybrid forecastmodel; electrical load forecasting; timeseries analysis; randomforest; multilayerperceptron 1. Introduction Recently, the smart gridhasbeengainingmuchattentionas a feasible solution to the current globalenergyshortageproblem[1]. Since ithasmanybenefits, includingthoserelatedtoreliability, economics, efficiency,environment,andsafety,diverse issuesandchallenges to implementingsucha smartgridhavebeenextensivelysurveyedandproposed[2].Asmartgrid [1,2] is thenext-generation powergrid thatmerges informationandcommunicationtechnology(ICT)with theexistingelectrical grid toadvanceelectricalpowerefficiencyto the fullestbyexchanging informationbetweenenergy suppliersandconsumers inreal-time[3]. Thisenables theenergysupplier toperformefficientenergy management forrenewablegenerationsources (solar radiation,wind,etc.) byaccurately forecasting power consumption [4]. Therefore, for amore efficient operation, the smart grid requires precise electrical loadforecasting inboth theshort-termandmedium-term[5,6]. Short-termloadforecasting (STLF)aimstoprepare for lossescausedbyenergyfailureandoverloadingbymaintaininganactive Energies2018,11, 3283;doi:10.3390/en11123283 www.mdpi.com/journal/energies119