资源描述：
《Wind Power Forecasts Using Gaussian Processes and NumericalWeather Prediction.pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、656IEEETRANSACTIONSONPOWERSYSTEMS,VOL.29,NO.2,MARCH2014WindPowerForecastsUsingGaussianProcessesandNumericalWeatherPredictionNiyaChen,ZhengQian,IanT.Nabney,andXiaofengMengAbstract—Sincewindattheearth’ssurfacehasanintrinsicallyapproachesforshort-termwindpowerforecasting:p
2、hysicalnu-complexandstochasticnature,accuratewindpowerforecastsaremericalweatherprediction(NWP)models,statisticalmodelsnecessaryforthesafeandeconomicuseofwindenergy.Inthisbasedpurelyonhistoricaldata,andstatisticalmodelswithNWPpaper,weinvestigatedacombinationofnumericand
3、probabilisticdataasadditionalexogenousinputs.Physicalmodelshavead-models:aGaussianprocess(GP)combinedwithanumericalweatherprediction(NWP)modelwasappliedtowind-powerfore-vantagesoverlongerhorizons(fromseveralhourstodozenscastinguptoonedayahead.First,thewind-speeddatafrom
4、NWPofhours),becausetheyinclude(3-D)spatialandtemporalfac-wascorrectedbyaGP,then,asthereisalwaysadefinedlimitontorsinafullfluid-dynamicsmodeloftheatmosphere.However,powergeneratedinawindturbineduetotheturbinecontrollingsuchmodelshavelimitations,suchasthelimitedobservations
5、etstrategy,windpowerforecastswererealizedbymodelingtheforcalibration(aseriousissuegiventheextremelylargenumberrelationshipbetweenthecorrectedwindspeedandpoweroutputusingacensoredGP.Tovalidatetheproposedapproach,threeofvariablesinthesemodels),therelativelylimitedspatialr
6、eso-real-worlddatasetswereusedformodeltrainingandtesting.lutionpossibleoversuchawidearea(typicallythewholeearth),Theempiricalresultswerecomparedwithseveralclassicalwindandtheimpossibilityofaccountingforlocaltopography[5].forecastmodels,andbasedonthemeanabsoluteerror(MAE
7、),Statisticalmodelswhichuseonlyhistoricalwindspeedandtheproposedmodelprovidesaround9%to14%improvementpowerdataanddonotincludeanyexplicitmodelofthephys-inforecastingaccuracycomparedtoanartificialneuralnetwork(ANN)model,andnearly17%improvementonathirddataseticalprocessesha
8、vebeendevelopedbasedonsinglemodelsorwhichisfromanewly-builtwindfarmforwhichthereisalimitedhybridsofseveralmeth
