Page - 164 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2038 donebymeansof crossvalidation. However, creatingagrid for all of theparameters tobe tuned impliesanextremelyhighcomputationalcost. 3. PredictionResults for theUniversityBuildings In this section, the fourensemblemethods thataredescribedaboveareappliedto theelectricity consumptionofasmall campusuniversity toevaluate theadequacyofeachtechnique in this typeof customers. Specifically,wewill focuson48-h-aheadpredictions inorder toapply themtothecontext ofDirectMarketConsumers,althoughdifferentpredictionhorizonswillbealsoconsideredfor the caseofXGBoostmethod. Someotheraspects, suchuspredictors importanceorparameterselection, foreachmethodarealsodeveloped. Firstly, in this section, thecustomer instudyis introduced. Secondly, the loaddata,predictors, andsomegoodnessoffitmeasurementsaredepicted. Finally, the forecastingresults for thecasestudy areshown. 3.1. CustomerDescription:ACampusUniversity Thecampus“AlfonsoXIII”of theTechnicalUniversityofCartagena (UPCT,Spain)comprises sevenbuildingsrangingfrom2000m2 to6500m2andameetingzone(10,000m2). Buildingsareof two kinds: naturallyventilatedcellular (individualwindows, local lightswitches,andlocalheatingcontrol) andnaturally ventilatedopen-plan (office equipment, light switched in longer groups, andzonal heating control). This campushas an overall surface larger than 35,500m2 to fulfill the needs of differentFaculties for classrooms,departmentaloffices, administrativeoffices, and laboratories for 1800studentsand200professors.Unfortunately, theageofbuildings (50yearsold in fourcases)and architectural conditioningworksare far fromactualenergyefficiencystandards, specifically in the two mainelectricalend-usesof thebuilding: air conditioning/spaceheating(lowperformance, insufficient heat insulation,andanimportantclusterof individualappliances forofficesandsmall laboratories) andlighting(whereconventionalmagneticballastsandfluorescentarestillusedatagreatextend). Withrespect to theshareofend-uses in the“CampusAlfonsoXIII”ofUPCT,heating,ventilation, andairconditioning(HVAC)is the largestenergyend-use(this trendis thesamebothintheresidential andnon-residentialbuildingsinSpainandothercountries,seeTable1)with40–50%ofoveralldemand; lightingfollowswith25–30%,electronicsandofficeequipment7–12%andotherapplianceswith8–10% (i.e., vendingmachines, refrigeration,water heatersWH, laboratory equipment, etc.). Notice that buildingtype iscritical inhowenergyendusesaredistributed ineachspecificbuilding. Table1shows acomparativeofend-uses inofficebuildings in threecountries [39]andin theanalysedcase, campus “AlfonsoXIII”. Table1.Energydemandinofficebuildingsbyend-use. End-Use USA(%) UK(%) Spain(%) UniversityBuildings (%) (UPCT) HVAC 48 55 52 40–50 Lighting 22 17 33 25–30 Equipment (appliances) 13 5 10 7–12 Other (WH,refrigeration) 17 23 5 8–10 3.2.DataDescription Dataused in thispaper correspond to the campusAlfonsoXIII of theTechnicalUniversityof Cartagena,asdescribedintheprevioussubsection.Hourly loaddatafrom2011to2016(bothincluded) wereanalyzed,obtainedfromtheretailerelectric companies (NexusEnergíaS.A.andIberdrolaS.A.). It iswell known that electricity consumption is related to several exogenous factors, such as the hourof theday, thedayof theweek, or themonthof theyear, and therefore these factorsmustbe taken intoaccount in thedesignof thepredictionmodel. Temperature isa factor thatmightaffect the electricityconsumption(coolingandheatingoftheuniversitybuildings). Thus, thehourlytemperature 164