基于TimeGAN-LSTM的月径流预测模型  

Monthly Runoff Prediction Based on TimeGAN-LSTM Model

作　　者：苗磊 刘杨[2] 田童晖 张志强 韩会明 MIAO Lei;LIU Yang;TIAN Tong-hui;ZHANG Zhi-qiang;HAN Hui-ming(Shaanxi Reconnaissance Design&Research Institute of Water Environmental Engineering,Xi’an 710018,China;School of Civil and Hydraulic Engineering,Huazhong Universityof Science and Technology,Wuhan 430074,China;Jiangxi Academy of Water Science and Engineering,Nanchang 230029,China)

机构地区：[1]陕西水环境工程勘测设计研究院,陕西西安710018 [2]华中科技大学土木与水利工程学院,湖北武汉430074 [3]江西省水利科学院,江西南昌230029

出　　处：《水电能源科学》2024年第11期12-15,共4页Water Resources and Power

基　　金：江西省自然科学基金资助项目(20232BAB213076);江西省水利科技项目(202224ZDKT06)。

摘　　要：准确的月径流预测对于水资源管理和配置具有重要意义,针对月径流样本数据不足导致预测精度较低的问题,提出了基于TimeGAN-LSTM的月径流预测模型。以无定河流域为研究区域,收集了6个气象站和4个水文站的数据资料,首先基于长短期记忆网络(LSTM)进行了4个站点的月径流预测,然后采用TimeGAN模型生成了月径流及其对应的月平均气温和降水数据,利用生成数据扩充了训练集并进行了基于TimeGAN-LSTM模型的训练和预测。研究结果表明,基于TimeGAN模型的生成数据与实测数据在分布上具有一定相似性;相较于LSTM,TimeGAN-LSTM模型预测结果的纳什效率系数平均增加了16.67%,平均绝对百分比误差降低了40.42%。该方法可有效提升月径流预测精度,为水资源管理决策提供技术支撑。Accurate monthly runoff prediction is of great significance for water resources management and allocation.To solve the problem of lower prediction accuracy due to insufficient monthly runoff sample data,a monthly runoff pre-diction model based on TimeGAN-LSTM was proposed.Taking Wuding River Basin as the research area,the data from 6 meteorological stations and 4 hydrological stations were collected.Firstly,the monthly runoff of 4 stations was predic-ted based on long short-term memory network(LSTM).And then,the monthly runoff and its corresponding monthly mean temperature and precipitation data were generated using the TimeGAN model.The generated data was used to ex-pand the training set and conduct training and prediction based on the TimeGAN-LSTM model.The results show that the distribution of monthly runoff forecast data generated by the TimeGAN model is similar to the measured monthly runoff data.Compared with the LSTM,the NSE predicted by the TimeGAN-LSTM model increased by 16.67%on average,and MAPE decreased by 40.42%.The proposed method can effectively improve the prediction accuracy of monthly runoff and provide technical support for water resources management decision-making.

关 键 词：月径流预测 无定河流域 TimeGAN LSTM 

分 类 号：TV121.4[水利工程—水文学及水资源] P338[天文地球—水文科学]