人民珠江2025，Vol.46Issue(11)：91-99,9.DOI:10.3969/j.issn.1001-9235.2025.11.011
基于积分时滞启发的深度学习水位预测模型
Deep Learning-Based Water Level Prediction Model Inspired by Integrator Delay
摘要
Abstract
Accurate water level prediction plays a key role in the safe operation of hydropower stations and is an important guarantee for improving power generation efficiency.The traditional deep learning method has limitations in dealing with the complex dynamics and time-delay characteristics of hydrological systems.It is difficult to capture the subtle changes and long-term and short-term dependencies in nonlinear hydrological processes,and its prediction accuracy cannot be guaranteed.Therefore,this study proposed a physically inspired hydraulic prediction model(PHM)to predict the water level in front of the single-channel pool dam of the Dadu River.The model deeply explored the internal mechanism of the physically inspired integrator delay model,mining the long-term dependence and short-term attribute correlation in hydrological data,effectively overcoming the problem of the deep learning model's insufficient ability to estimate time delay dynamics,and improving the accuracy of prediction results.In this paper,we used the real data of the Dadu River.Through a large number of experiments,it is proven that compared with the existing models,PHM has a significant reduction in the three key indicators of mean absolute error(MAE),root mean square error(RMSE),and mean absolute percentage error(MAPE).The average reduction of MAE is 43.2%;the average reduction of MAPE is 42.9%,and the average reduction of RMSE is 52.3%.The experimental results show that the model is more practical and reliable than the existing models in different scenarios.关键词
积分时滞/水位预测/深度学习/物理启发/单渠池Key words
integrator delay/water level prediction/deep learning/physical inspiration/single-channel reservoir分类
建筑与水利引用本文复制引用
罗玮,吴佳豪,陈灵强,郑钧,雷晓辉,王立志,谭敏,许江涛..基于积分时滞启发的深度学习水位预测模型[J].人民珠江,2025,46(11):91-99,9.基金项目
河北省自然科学基金资助(E2024402142) （E2024402142）
