南水北调与水利科技（中英文）2026，Vol.24Issue(4)：965-976,12.DOI:10.13476/j.cnki.nsbdqk.2026.0090
因果分析-深度学习耦合的南水北调中线干渠水质预测方法
Causality analysis-deep learning based water quality analysis and prediction for the South-to-North Water Transfers Project
摘要
Abstract
The South-to-North Water Transfer Project(SNWTP)is a major strategic initiative designed to mitigate water scarcity in northern China.As a long-distance artificial channel,the SNWTP's ecological system has not yet reached a stable state,posing potential risks to water quality and aquatic ecology,particularly in sections susceptible to algal blooms.Effective monitoring and prediction of water quality parameters are crucial for maintaining water safety and sustainability.A hybrid method combining empirical dynamic modeling(EDM)with deep learning was proposed to analyze and forecast water quality at three vulnerable monitoring sections along the Middle Route of the SNWTP,which are at risk of algal outbreaks.Causal relationships among key water quality indicators were first examined using EDM.Subsequently,the identified causal variables were incorporated into the construction of predictive datasets,which were then used to train deep neural network models for water quality prediction.The application of EDM revealed bidirectional causal relationships between total phosphorus and both pH and dissolved oxygen,as well as between total nitrogen and the same two parameters,within the studied sections.By integrating these causality-based indicators into the predictive framework,the deep learning models achieved high forecasting accuracy,with Nash-Sutcliffe efficiency coefficients exceeding 0.9.The proposed method demonstrated superior performance in capturing dynamic interactions among water quality variables compared with conventional correlation-based analyses.Unlike traditional correlation analysis,EDM effectively elucidated causal linkages among water quality indicators and provided deeper insight into their interdependent behavior.The combined EDM-deep learning approach offers a reliable and accurate tool for water quality prediction and supports subsequent analysis and early warning of algal bloom dynamics in large-scale water diversion projects.关键词
因果分析/预测/经验动态建模/深度学习/水质Key words
causal analysis/prediction/empirical dynamic modeling/deep learning/water quality分类
建筑与水利引用本文复制引用
白冰,董飞,彭文启,刘晓波..因果分析-深度学习耦合的南水北调中线干渠水质预测方法[J].南水北调与水利科技（中英文）,2026,24(4):965-976,12.基金项目
National Key R&D Program of China(2021YFC3200903) （2021YFC3200903）
the Basic Scientific Research Expense Project of IWHR(WE0145B032021) （WE0145B032021）
