Page 115 - 水利学报2025年第56卷第3期
P. 115
shorttermmemorynetworks[J].JournalofHydrology:RegionalStudies,2022.doi:10.1016?j.ejrh.2022.101034.
[18] SUDERSHANG,DANL,SHIHCHIEH K,etal.Machinelearningassistedreservoiroperationmodelforlong -
term watermanagementsimulation[J].JournaloftheAmericanWaterResourcesAssociation,2022,58(6):
1592 - 1603.
[19] 田世民,江恩慧,王远见,等.基于黄河流域系统治理的水库多目标调度约束阈值研究[J].水利学报,
2024,55(6):631 - 642,665.
[20] 王天宇,董增川,付晓花,等.黄河上游梯级水库防洪联合调度研究[J].人民黄河,2016,38(2):40 - 44.
[21] 于显亮,彭杨,李颖曼,等.黄 河 上 游 梯 级 水 库 汛 期 增 泄 联 合 调 度 研 究 [J].人 民 黄 河,2023,45(8):
68 - 72,78.
[22] 刘龙庆,刘玉环,张献志,等.龙羊峡水库后汛期入库径流特征及可 蓄 水 量 分 析 [J].人 民 黄 河,2024,
46(2):38 - 40,48.
[23] HOCHREITERS,SCHMIDHUBERJ.Longshort - termmemory[J].NeuralComput,1997,9(8):1735 - 1780.
[24] KINGMADP,BAJ.Adam:Amethodforstochasticoptimization[J].ComputerScience,2014.doi:10.48550?
arXiv.1412.6980.
[25] ABADIM,BARHAM P,CHENJ,etal.TensorFlow:asystem forlarge - scalemachinelearning[C]??Proceed
ingsofthe12thUSENIXconferenceonOperatingSystemsDesignandImplementation.2016.
[26] 王战策,谢小平,曹光明.龙羊峡水库径流调节作用及效益分析[J].人民黄河,2017,39(1):14 - 17.
[27] 舒鹏,熊立华,陈杰,等.基于目标库容曲线的水库出流模拟模型[J].水利学报,2023,54(11):1323 -
1333,1346.
SimulationofreservoiroutflowunderinconsistentconditionbasedonMDUPLEX - LSTM
1
1
1
YUXinpeng,WUShuyue,ZHAOJianshi
(DepartmentofHydraulicEngineering,TsinghuaUniversity,Beijing 100084,China)
Abstract:Forthehydrologicalsimulationofriverbasinswithreservoirs,accuratelysimulatingreservoirs’outflow
iscrucialfortheaccuratesimulationofstreamflow.Undertheinconsistencyamongclimaticandhydrologicalcondi
tionsaswellashumanactivities ,itisstillachallengehowtoreasonablysimulatethereservoirdischargeprocess.
Inthispaper ,bothLongyangxia,amulti - yearregulationreservoir,andLiujiaxia,anannualregulationreservoir,
intheupperreachesoftheYellowRiveraretakenasresearchobjects.AreservoiroutflowsimulationmethodMDU
PLEX - LSTM,whichcouplesdeterministicdatasetsamplingalgorithmandmachinelearning,isproposedandes
tablished.TheresultsshowthattheMDUPLEX - LSTM canlearntheactualreservoiroperationpatternsunderdif
ferenthydrologicalconditions.MDUPLEX - LSTM achievedNSEvaluesofover0.84and0.73forthedailyoutflow
simulationsoftheLongyangxiaandLiujiaxiareservoirs ,respectively.Additionally,thediscrepancyofthesimula
tionaccuracyamongthecalibrationtestandvalidationdatasetsisnegligible ,indicatingtherobustnessofMDU
PLEX - LSTM.
Keywords:simulationofreservoiroutflow;MachineLearning;LSTM;MDUPLEX;reservoiroperation;upper
YellowRiver
(责任编辑:耿庆斋)
— 3 9 7 —