Page 84 - 2023年第54卷第11期
P. 84
WRF?WRF - Hydrocoupledstreamflowforecastingbasedonreal - timeupdatingusingLSTM
1
2
2
1,2
1
LIUYuchen ,LIUJia,LIULusan,LIChuanzhe,WANGYu
(1.ChineseResearchAcademyofEnvironmentalSciences,Beijing 100012,China;
2.StateKeyLaboratoryofSimulationandRegulationofWaterCycleinRiverBasin,
ChinaInstituteofWaterResourcesandHydropowerResearch,Beijing 100038,China)
Abstract:InordertoimprovetherunoffpredictionperformanceoftheWRF?WRF - Hydrocoupledatmospheric -
hydrologicsystemsandreducetheerrorsinpeaktimeandfloodpeakflowprediction ,thisstudyusesvariationalda
taassimilationtechnologytoreducetherainfallpredictionerror,atthesametime,areal - timecorrectionstudyon
therunoffpredictionprocessoftheWRF?WRF - Hydrosystem isconductedusingthelongshort - term memory
(LSTM),andcomparethereal - timecorrectionresultswiththeautoregressivemovingaveragemodel(ARMA).
TheresearchresultsindicatethatdataassimilationtechnologycaneffectivelyimprovetheaccuracyofWRFmodel
rainfallpredictionandreducetheinputerrorofWRF - Hydromodel ,buttheaccuracyofrunoffpredictionstill
needstobeimproved.Comparingthereal - timecorrectionresultsofLSTM andARMAmodelsforrunoffforecas
ting ,itwasfoundthatduringthefirstthreehoursoftheforesightperiod,theperformanceofthetwomodelsisba
sicallysimilarofsmallwatershedsinsemihumidandsemi - aridmountainousareasinnorthernChina.Exceptfor
Event4,theattenuationratesofLSTMandARMAmodelsinthethreehoursoftheforesightperiodare2.04~23.08
and9.18~36.47 ,respectively.Astheforesightperiodextends,thedecayrateofLSTM runoffpredictionaccuracy
isgenerallyslowerthantheARMAmodel ,andthepredictioneffectisbetterthantheARMAmodel.
Keywords:LSTM;real - timeupdating;WRF?WRF - Hydro;runoffforecast;dataassimilation
(责任编辑:韩 昆)
(上接第 1333页)
Reservoiroutflowsimulationmodelbasedonreservoirtargetcapacitycurve
SHUPeng,XIONGLihua,CHENJie,YINJiabo,CHENGLei,ZHOUYanlai
(StateKeyLaboratoryofWaterResourcesandHydropowerEngineeringScience,WuhanUniversity,Wuhan 430072,China)
Abstract:Withthecontinuousconstructionofreservoirs,theimpactofreservoirsonthewatercycleismoreand
moresignificant ,andtheoutflowprocessofreservoirshasbecomeanimportantpartofthelandwatercycle.In
ordertosimulatetheimpactofReservoironwatercirculation ,ReservoirOutflow SimulationModel(ROSM)
shouldbeconstructedtosimulatetheoutflowofreservoir.However,thecurrentproposedROSMshaveweakgen
eralityandareonlyapplicabletoreservoirswithpresetoperationmode.Inthispaper ,itisassumedthatthegoal
reservoircapacityatvarioustimesoftheyearformsaTargetCapacityCurve (TCC).ThegeneralityofROSM can
beimprovedbysettingdifferentTCCsfordifferentreservoirs.Twomodelstructuresareproposedtoconsiderthe
influenceofTCConreservoirdischargefromtwoaspects :(1)Thereservoirtendstostorewaterandtheoutflow
decreasesintheseasonwithhighertargetstoragecapacity,whilethereservoirtendstoreleasewaterandtheout
flowincreasesintheseasonwithlowertargetstoragecapacity.(2)Theoutflowofthereservoirshouldmakethe
actualwaterstorageclosetothetargetstoragecapacity.Accordingtothetwomodelstructures,fourkindsofTCC
- basedROSM wereconstructed ,andthecommonROSM wereappliedto26reservoirsintheYangtzeRiver,
PearlRiverandYellowRiverbasins.Theresultsshowedthat:(1)theperformanceoftheoutflowsimulationand
waterstoragesimulationoftheROSM consideringthetwomodelstructureswasoptimal ,indicatingthatthestruc
tureofthetwomodelswasreasonableandeffective ;(2)ConsideringthatthesimulationaccuracyofTCCROSM
inmulti - yearandannualregulationreservoirsisfarhigherthanthatofotherROSM ,itindicatesthatTCCcanef
fectivelyimprovethesimulationaccuracyofthesetwotypesofreservoirsandimprovetheversatilityofROSM.
Keywords:reservoiroutflow;simulationmodel;reservoirstorage;reservoirtargetcapacitycurve
(责任编辑:于福亮)
— 1 3 6 —
4