基于长短时记忆网络的顶托影响下干支流洪水模拟研究  

Research on Flood Simulation Considering Backwater Effect Using Long Short-term Memory Networks

作　　者：张艺佳 吴剑[1,3] 彭勇[1,4] 丁勇 郭家园[5] ZHANG Yi-jia;WU Jian;PENG Yong;DING Yong;GUO Jia-yuan(School of Infrastructure Engineering,Dalian University of Technology,Dalian 116024,China;Hengshui Hydrological Survey and Research Center of Hebei Province,Hengshui 053000,China;School of Chemical Engineering,Ocean and Life Sciences,Dalian University of Technology,Panjin 124221,China;Ningbo Institute of Dalian University of Technology,Ningbo 315000,China;Songliao Water Resources Committee of the Ministry of Water Resources,Changchun 130000,China)

机构地区：[1]大连理工大学建设工程学院,辽宁大连116024 [2]河北省衡水水文勘测研究中心,河北衡水053000 [3]大连理工大学化工海洋与生命学院,辽宁盘锦124221 [4]大连理工大学宁波研究院,浙江宁波315000 [5]水利部松辽水利委员会,吉林长春130000

出　　处：《水电能源科学》2024年第10期24-28,共5页Water Resources and Power

基　　金：国家重点研发计划(2022YFC3202803);国家自然科学基金项目(52309009)。

摘　　要：干支流交汇河段易发生洪水相互顶托现象,造成河段持续处于高水位,极大地增加了洪水模拟的难度。以三岔河口上游受顶托影响显著的嫩江大赉站为研究对象,首先分析识别大赉站历史洪水的顶托关系,并根据洪水顶托关系划分洪水类型;在此基础上采用长短时记忆(LSTM)网络建立洪水模拟模型,评估模型的模拟效果。结果表明,采用流量、水位变化率可以较为有效地识别洪水顶托关系,历史上嫩江受到洪水顶托影响的年份较多;LSTM模型输入中仅考虑上游来水对大赉站流量模拟精度影响相对较小,而对水位模拟精度影响显著;考虑顶托影响的LSTM模型对大赉站的流量、水位模拟精度均较高。可见,所构建的LSTM模型能较准确地模拟出顶托影响下的大赉站洪水过程,为类似流域或站点的洪水模拟提供参考。Flood backwater generally occurs where main and branch rivers meet,resulting in a prolonged high water level and posing significant challenges to flood simulation.This paper focuses on Dalai station,located the downstream of Nenjiang River,which is obviously affected by flood backwater.Firstly,the backwater types of historical flood events recorded at Dalai Station was analyzed.And then these flood events were categorized based on the identified backwater types.On this basis,a flood simulation model was developed using long short-term memory(LSTM)networks,and its performance was further evaluated.The results show that the change rate of observed flow and water level can effectively identify the types of backwater,and the Nenjiang River experiences the backwater effect in most years.Meanwhile,for the LSTM model input,considering only upstream inflow has a relatively minor impact on the accuracy of simulated flow,while it significantly affects the accuracy of simulated water level.The LSTM model obtains high accuracy in both simulated flow and water level when considering downstream backwater effects.The developed LSTM model accurately captures the flood dynamics at the selected Dalai station affected by downstream backwater effect,which offers reference for flood simulation in similar basins or stations.

关 键 词：机器学习 长短时记忆网络 洪水顶托 洪水模拟 

分 类 号：TV122.5[水利工程—水文学及水资源] P333.2[天文地球—水文科学]