水生态学杂志2026，Vol.47Issue(3)：172-178,7.DOI:10.15928/j.1674-3075.202401290034
机器学习在水库下泄水温预测中的应用分析
Application of Machine Learning to Predict the Discharge Water Temperature of Reservoirs
摘要
Abstract
The operation of hydropower station significantly alters the temporal and spatial distribution of water temperature in the river channel downstream of the dam.In this study,Pubugou hydropower sta-tion on Dadu River in the upper reaches of Yangtze River was selected for research,and we explored the reliability of five machine learning models in predicting the temperature of water discharged from the res-ervoir.Our aim was to develop a method for high-precision water temperature prediction and to provide technical support for optimizing the ecological operation and management of reservoirs.The five learning models included the Random Forest(RF),Support Vector Regression(SVR),Light Gradient Boosting Machine(LGBM),Convolutional Neural Network(CNN),and Long Short-Term Memory Network(LSTM).To begin with,the Spearman correlation coefficient was used to screen meteorological factors and reservoir operation parameters that affect the temperature of water discharged from the reservoir.Then,a genetic algorithm was applied to optimize the parameters of the five models used to predict dis-charge water temperature.Results show:(1)The dew point temperature exhibited the highest correlation with the discharge water temperature,with a correlation coefficient of 0.89,while the correlation coeffi-cients between wind speed,cloud cover,solar radiation and reservoir water level with discharge water temperature were all less than 0.4.(2)The models optimized by the genetic algorithm all performed well with the training set,but the RF model gave the best results(r2=0.997),while the LSTM model gave the worst results(r2=0.953).(3)In predicting discharge water temperature,all models gave a good fit,with r2＞0.931,a mean absolute error ≤0.662 ℃,and a mean square error ≤0.852 ℃.Among them,the RF and LGBM models had narrow residual ranges,and the maximum residuals of the SVR and CNN models were smaller.In conclusion,the machine learning methods optimized using the genetic algorithm can effec-tively predict the temperature of water discharged from reservoirs.关键词
水库下泄水温/水温预测/机器学习/遗传算法/模型优化Key words
reservoir discharge water temperature/water temperature prediction/machine learning/genetic algorithm/model optimization分类
建筑与水利引用本文复制引用
陈俊光,杨世伟,王远铭,梁瑞峰,李克锋..机器学习在水库下泄水温预测中的应用分析[J].水生态学杂志,2026,47(3):172-178,7.基金项目
科技基础资源调查专项(2022FY100203) （2022FY100203）
国家自然科学基金联合基金重点项目(U2240212). （U2240212）
