基于LSTM和BP神经网络的水库入库径流中长期预测比较研究  

Comparative Study on Mid-Long Term Prediction of Reservoir Inflow based on LSTM and BP Neural Network

作　　者：邹红梅 朱成涛 ZOU Hongmei;ZHU Chengtao(Hydrology Bureau of Changjiang Water Resources Commission,Wuhan 430010,China;Yalong River Hydropower Development Company.Ltd.Chengdu 610051,China)

机构地区：[1]长江水利委员会水文局,湖北武汉430010 [2]雅砻江流域水电开发有限公司,四川成都610051

出　　处：《水文》2024年第4期27-31,37,共6页Journal of China Hydrology

摘　　要：为探究不同数据驱动模型在中长期径流预报的应用效果,以雅砻江流域两河口水库、锦屏一级水库入库径流为研究对象,采用长短期记忆网络(LSTM)和BP神经网络模型对各水库年径流、月径流进行预测。基于前期径流信息和环流影响因子数据构建中长期径流预测因子集,对长短期记忆网络(LSTM)和BP神经网络参数进行优选,建立各水库的年、月径流预测模型。结果表明:两种模型在年径流和月径流预测精度比较高,长短期记忆网络(LSTM)在年、月径流预测精度都略高于BP神经网络模型,两种模型在两河口水库径流预测精度都高于锦屏一级水库。该研究成果可为大型水电站中长期径流预测提供借鉴。In order to explore the application effect of different data-driven models in mid-long term runoff prediction,taking the inflow runoff of Lianghekou Reservoir and Jinping First Class Reservoir in the Yalong River Basin as the research objects,the long short-term memory neural network(LSTM)and BP neural network models were used to predict the annual and monthly run⁃off of each reservoir.A set of medium and long term runoff prediction factors was constructed based on previous runoff informa⁃tion and circulation impact factor data.The parameters of LSTM and BP neural network were optimized,and annual and monthly runoff prediction models for each reservoir were established.The prediction results show that the LSTM has higher accuracy in predicting annual and monthly runoff than the BP neural network model,and both models have higher accuracy in predicting run⁃off in the Lianghekou Reservoir than in the Jinping First Class Reservoir.The results can provide references for mid-long term runoff prediction of large hydropower stations.

关 键 词：LSTM BP神经网络 中长期径流预测 雅砻江流域 

分 类 号：P338.2[天文地球—水文科学] TV121.4[水利工程—水文学及水资源]