Seite - 229 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article ScalableClusteringofIndividualElectricalCurves forProfilingandBottom-UpForecasting BenjaminAuder1, JairoCugliari 2,∗ ID ,YannigGoude3 andJean-MichelPoggi 4 1 LMO,UniversityParis-Sud,91405Orsay,France;benjamin.auder@math.u-psud.fr 2 ERICEA3083,UniversitydeLyon,Lyon2,69676Bron,France 3 EDFR&D,LMO,UnivParis-Sud,91405Orsay,France;yannig.goude@edf.fr 4 UniversityParisDescartes&LMO,Univ. Paris-Sud,91405Orsay,France; jean-michel.poggi@math.u-psud.fr * Correspondence:jairo.cugliari@univ-lyon2.fr;Tel.: +33-4-7877-3155 Received: 29 June2018;Accepted: 16 July2018;Published: 20 July2018 Abstract: Smart grids require flexible data driven forecastingmethods. We propose clustering tools forbottom-upshort-termloadforecasting.Wefocusonindividualconsumptiondataanalysis whichplaysamajor role forenergymanagementandelectricity loadforecasting. Thefirst section is dedicatedto the industrial contextandareviewof individualelectricaldataanalysis. Then,wefocus onhierarchical time-series forbottom-upforecasting. The idea is todecompose theglobal signaland obtaindisaggregatedforecasts insuchawaythat their sumenhances theprediction. This isdone in threesteps: identifyarather largenumberofsuper-consumersbyclustering theirenergyprofiles, generateahierarchyofnestedpartitionsandchoose theone thatminimizeapredictioncriterion. Using a nonparametricmodel to handle forecasting, andwavelets to define various notions of similaritybetween loadcurves, thisdisaggregationstrategygivesa16%improvement in forecasting accuracywhenappliedtoFrenchindividualconsumers. Then, thisstrategy is implementedusing R—thefreesoftwareenvironment forstatistical computing—sothat it canscalewhendealingwith massivedatasets. Theproposedsolution is tomake thealgorithmscalable combinedata storage, parallel computinganddoubleclusteringsteptodefinethesuper-consumers. Theresultingsoftware isopenlyavailable. Keywords: clustering; forecasting;hierarchical time-series; individualelectrical consumers; scalable; short term;smartmeters;wavelets 1. Introduction 1.1. IndustrialContext Energy systems are facing a revolution andmany challenges. On the one hand, electricity production ismovingtomore intermittencyandcomplexitywith the increaseof renewableenergy andthedevelopmentof smalldistributedproductionunits suchasphotovoltaicpanelsorwindfarms. Ontheotherhand,consumption isalsochangingwithplug-in (hybrid)electricvehicles,heatpumps, the development of new technologies such as smart phones, computers, robots that often come withbatteries. Tomaintain the electricityquality, energy stakeholders aredeveloping smart grids (see [1,2]), thenextgenerationpowergrid includingadvancecommunicationnetworksandassociated optimisationandforecastingtools.Akeycomponentof thesmartgridsaresmartmeters. Theyallow two-sidedcommunicationwith thecustomers, real timemeasurementof consumptionanda large scopeofdemandsidemanagementservices.Alotofcountrieshavedeployedsmartmeters, as stated in [3], theUK, theUSandChinahaverespectivelydeployed2.9,70and96millionofsuchequipments in2016. InFrance,35millionwillbedeployedbefore2021foraglobalcostof5billion(seee.g., [4]). Energies2018,11, 1893;doi:10.3390/en11071893 www.mdpi.com/journal/energies229