基于MDUPLEX-LSTM的非一致条件下的水库出流模拟  

Simulation of reservoir outflow under inconsistent condition based on MDUPLEX-LSTM

作　　者：余昕鹏 吴书悦 赵建世 YU Xinpeng;WU Shuyue;ZHAO Jianshi(Department of Hydraulic Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学水利水电工程系,北京100084

出　　处：《水利学报》2025年第3期387-397,共11页Journal of Hydraulic Engineering

基　　金：国家自然科学基金项目(黄河水科学研究联合基金)(U2243215);河南省黄河流域生态环境保护与修复重点实验室开放基金项目(LYBEPR202404)。

摘　　要：对于建有水库的流域开展水文模拟时,合理模拟水库出流过程对于流域径流的准确模拟至关重要。在气候水文条件和人类活动非一致的状况下,如何合理模拟水库出流过程有待深入探索和研究。本文以黄河上游的多年调节水库龙羊峡水库和年调节水库刘家峡水库为研究对象,提出了耦合确定性数据集划分算法与机器学习的MDUPLEX-LSTM水库调度模拟方法。结果表明,MDUPLEX-LSTM能够学习历史时期不同水文情势下的水库实际调度规律,采用该方法模拟龙羊峡和刘家峡两座水库的日尺度出流过程,NSE分别达到0.84和0.73,模拟精度高,且在率定、测试、验证集上的模拟精度非常接近,鲁棒性高。For the hydrological simulation of river basins with reservoirs,accurately simulating reservoirs’outflow is crucial for the accurate simulation of streamflow.Under the inconsistency among climatic and hydrological conditions as well as human activities,it is still a challenge how to reasonably simulate the reservoir discharge process.In this paper,both Longyangxia,a multi-year regulation reservoir,and Liujiaxia,an annual regulation reservoir,in the upper reaches of the Yellow River are taken as research objects.A reservoir outflow simulation method MDUPLEX-LSTM,which couples deterministic dataset sampling algorithm and machine learning,is proposed and established.The results show that the MDUPLEX-LSTM can learn the actual reservoir operation patterns under different hydrological conditions.MDUPLEX-LSTM achieved NSE values of over 0.84 and 0.73 for the daily outflow simulations of the Longyangxia and Liujiaxia reservoirs,respectively.Additionally,the discrepancy of the simulation accuracy among the calibration test and validation datasets is negligible,indicating the robustness of MDUPLEX-LSTM.

关 键 词：水库出流模拟 机器学习 LSTM MDUPLEX 水库调度 黄河上游 

分 类 号：TV691[水利工程—水利水电工程]