西班牙可再生能源中心风功率预测

PedroCorreia:[e-mail]ías:[e-mail]ánMoya:[e-mail]edictionmodels(NWP),havebeentraditionallyusedtopredicttherealstateofearth'satmosphere.Theinitialstateoftheatmosphere(ANALYSIS)isreproducedusingmeasurementsfromsatellites,weatherstations,etc,andconvertedtoaregulargridthatcanbeusedtofeedthemesoescalemodel.Byresolvingtheprimitiveequationswiththatinputdata,theNWPmodelscanpredicttheweatherinthefuture.There'sdifferenttypesofatmosphericmodels:Global:GlobalForecastingsystem(GFS)fromNcep/NCARandtheECMWFGlobalmodel,...Regional:Skiron;EtaModel;WRF;MM5,Hirlam;Aladin,...MesoescalemodelthatusesthebaseoftheETAmodel.RequiresanUNIXOperatingSystem;Itisabletousetheweatherinputdatafrom;GFS(GlobalForecastingSystem);NCEP/NCARReanalysis1;ECMWF(GlobalModel)‏CENERworkswithSKIRONsinceOctober,2005.CENERworkswithSKIRONsinceOctober,2005.Itwasfirstconfiguredtorunreal-timeforecasts,allowingCENERtoobtaindailyweatherpredictions.Fromsometimenow,themodelisalsobeenusedtowindanddirectsolarradiationresourceassessmentinanwiderangeoflocationsthroughouttheglobe.Togeneratewind/solarradiationmaps,it'sdesirabletorunSKIRONaslongaspossible(morethan5years),inordertoobtainthelongtermbehaviorofthedifferentmeteorologicalvariables,suchaspressure,windvelocityanddirection(atseveralheightsabovegroundlevel),directsolarradiation,temperature,etc.SKIRON:Real-timepredictionsIt'sexecutedin16processorswithanhorizonof180ph-5h30minHorizontalResolution:0.1ºx0.1º(~10kmx10km)-341x281ptsVerticalResolution:38Etaverticallevels.NonestingTemporalresolution:outputfrequency=1h(180dailyfiles)Dailydownloadandstorageof:GFS12UTC,SST,SnowcoverandSnowdepthBackupsystem:Thesamemodelconfiguration,indifferentmachineswhicharelocatedinanotherarea.Thisallowsustoguaranteetheclientsforecastsincaseofafailureinthemainsystem(powerfailure,computermalfunction,networkproblems,etc).SKIRON:Real-timeReal-timedomainfromOctober2005untilNovember2009.DomainsinceNovember2009untilnow.Intheelectricitygridatanymomentbalancemustbemaintainedbetweenelectricityconsumptionandgeneration-otherwisedisturbancesinpowerqualityorsupplymayoccur.Windgenerationisadirectfunctionofwindspeedand,incontrasttoconventionalgenerationsystems,isnoteasilydispatchable.Fluctuationsofwindgenerationthusreceiveagreatamountofattention.Managingthevariabilityofwindgenerationisthekeyaspectassociatedtotheoptimalintegrationofthatrenewableenergyintoelectricitygrids.Statisticalpredictionmethodsarebasedononeorseveralmodels(linearandnon-linear)thatestablishtherelationbetweenhistoricalvaluesofpower,aswellashistoricalandforecastvaluesofmeteorologicalvariables,andwindpowermeasurements.Modelparametersareestimatedfromasetofpastavailabledata,andtheyareregularlyupdatedduringonlineoperationbyaccountingforanynewlyavailableinformation(i.e.meteorologicalforecastsandpowermeasurements)..LocalPredisoperationalsince2001andhasbeencontinuouslydevelopedsincethen.Reliabilityandaccuracyarethemaincharacteristicsofthesystem.Reliabilityisbasedontheredundancyof:Hardware.Inputdata.Processes.Accuracyisobtainedthroughthecombinationofforecastswithdifferentinformation:“multi-modelensemble”.SupportVectorMachinetechnology,PCAalgorithms,dataqualitycontrol.Forecastsforoffshorewindfarms:Windfarmenergyproduction.Waves(WAM4highresolutionwaveforecasts).reducedvisibility.GFSSKIRONECMWFPCAMOSENSEMBLEMOS2DELIVERYFTPCENERFTPCLIENTFTPAGENTSVMLocalPredincludesacombinationalgorithmdevelopedincollaborationwithDTU-IMM.Thelevelofimprovementdependsontheerroroftheindividualforecastsandonthelevelofcorrelationbetweenthem.Thecombinationisabletoimprovethebestindividualforecast.TheactualSpanishelectricalmarket,allowsustocorrectthewindenergyforecastingpresentedinthedailymarketbymeansoftheintradailymarket.Thismarketisorganizedintosixsessionsandagentsthathavepreviouslyparticipatedinthedailymarketcanpresentnewprogramofproduction.Thenewpredictionsmustbepresentedbetweentheopeningandclosinghoursofthesession.Thus,ineachintradailysession,wecorrectamaximumoffivepredictions,andtakingintoaccounttheclosinghourofthesession,wecanusefromfourthtoeighthstepahead.Thereforetheimportanceoftheshorttimeforecastandsotheneedofashorttimemodelfallshere.Anewmodelforshort-termpredictionhasbeendevelopedtakingintoaccounttheSpanishmarketrules.Thismodelisfocusedinshortforecastinghorizons.First,itusesonlinepowerproductiondataofthewindfarmstobuilddifferenttimeseriesmodels(Box-JenkinsmethodologyandaversionofHoltWintersAlgorithm).Ontheotherhand,itutilizesexistingforecastsforthedailymarketproducedbyCENER’sLocalPredmodelbasedonmesoscaleNWPandMOScorrections.Finallyitimplementsacombinationalgorithmthatofferstheoptimalforecastforeachhorizon.WepresenttheresultsobtainedfromtheCENERshorttimemodelappliedonamediumwindfarmfromSpainbetweenFebruaryandDecember.Wepresenttheimprovementofthenewmodelagainstthepersistenceasshorttimepredictionandagainstthedailymarketforecasting.R+DEuropeanprojects(VIandVIIFrameworkProgram):UPWIND“Findingdesignsolutionsforverylargewindturbines”POW’WOW“PredictionOfWaves,W