长江科学院院报 ›› 2024, Vol. 41 ›› Issue (12): 9-14.DOI: 10.11988/ckyyb.20230804

• 水资源 • 上一篇    下一篇

基于深度学习的三峡电站未来坝前最大最小水位预测

王永强1,2,3(), 张森1,2,3, 谢帅1,2,3(), 周涛1,2,3   

  1. 1 长江科学院 水资源综合利用研究所,武汉 430010
    2 长江科学院 流域水资源与生态环境科学湖北省重点实验室,武汉 430010
    3 长江水利委员会 长江经济带保护与发展战略研究中心,武汉 430010
  • 收稿日期:2023-07-24 修回日期:2023-09-18 出版日期:2024-12-01 发布日期:2024-12-01
  • 通信作者:
    谢 帅(1993-),男,河南南阳人,工程师,博士,研究方向为水文预报。E-mail:
  • 作者简介:

    王永强(1982-),男,河南郑州人,正高级工程师,博士,研究方向为水电能源优化运行。E-mail:

  • 基金资助:
    国家重点研发计划重点专项(2022YFC3202300); 国家自然科学基金面上项目(42271044); 水利部重大科技项目(SKS-2022120); 湖北省自然科学基金联合基金项目(2022CFD027); 中央级公益性科研院所基本科研业务费项目(CKSF2021486); 中国长江电力股份有限公司资助项目(Z242302057)

Predicting Maximum and Minimum Future Water Levels in front of Three Gorges Dam Using Deep Learning

WANG Yong-qiang1,2,3(), ZHANG Sen1,2,3, XIE Shuai1,2,3(), ZHOU Tao1,2,3   

  1. 1 Water Resources Department, Changjiang River Scientific Research Institute, Wuhan 430010, China
    2 Hubei Provincial Key Laboratory of Basin Water Resources and Eco-environmental Sciences,Changjiang RiverScientific Research Institute,Wuhan 430010,China
    3 Research Center on Protection and Development Strategyfor Yangtze River Economic Belt, Changjiang Water Resources Commission, Wuhan 430010, China
  • Received:2023-07-24 Revised:2023-09-18 Published:2024-12-01 Online:2024-12-01

摘要:

最大最小水位是计算梯级水库调度问题、水电站经济运行问题等时要考虑的重要约束条件,其精确预测能为水电站经济运行提供支持。常用的迭代计算容易误差积累,导致多时段预测结果可信度降低。选取对时间序列问题有良好处理效果的长短时记忆网络模型(LSTM),将其应用于三峡电站未来4 d最大最小坝前水位预测中,依据水量平衡预测框架构建传统预测模型;基于LSTM使用不同特征变量构建2种深度学习模型,并比较其预测效果。计算结果表明,考虑三峡库区水面线传播规律后的深度学习模型预测具有精确稳定的预测效果,99%预测绝对误差<40 cm。

关键词: 水电站经济运行, 水位预测, LSTM, 深度学习, 神经网络, 三峡电站

Abstract:

The maximum and minimum water levels are crucial constraints in the calculation of cascade reservoir operations and the economic operation of hydropower stations. Traditional iterative methods for multi-period predictions lack credibility due to error accumulation. This study employs a Long Short-Term Memory (LSTM) model which is effective in handling time series problems to predict the maximum and minimum water levels of the Three Gorges Reservoir over the next four days. Two LSTM-based deep learning models incorporating different characteristic variables are developed, and a conventional forecast model based on the water balance framework is also constructed for comparison. Results demonstrate that the deep learning model, which considers the propagation law of water surface profiles in the Three Gorges Reservoir area, delivers accurate and stable predictions, achieving an absolute error of less than 40 cm for 99% of the predictions.

Key words: economic operation of hydropower station, water level prediction, LSTM, deep learning, neural network, Three Gorges hydropower station

中图分类号: