[Water Quality Analysis and Prediction for the Middle Route of South-to-North Water Diversion Project Based on EDM-LSTM].

To deeply analyze the causal relationships among various water quality indicators in the Middle Route of South-to-North Water Diversion Project and achieve high-precision predictions, a method combining empirical dynamic modeling （EDM） and deep learning is proposed. Empirical dynamic modeling is utilized to conduct causal analysis among water quality indicators. Based on this, a dataset is constructed to train long short-term memory （LSTM） neural networks for water quality prediction. The prediction accuracy and computational time of different LSTM structures are compared. The results showed that： ① The water quality of the Middle Route of South-to-North Water Diversion was stable, with no significant abrupt changes along the route. ② There was a bidirectional causal relationship between total nitrogen and dissolved oxygen, as well as pH, in the Middle Route of South-to-North Water Diversion Project. ③ The neural network trained based on causal analysis results could achieve high-precision water quality predictions for the Middle Route of South-to-North Water Diversion Project, with the Nash efficiency coefficient of the predictions generally exceeding 0.85. This method can deeply analyze the causal relationships among variables and achieve high-precision predictions, providing scientific support for water quality management and subsequent analysis and prediction of water ecological factors in the Middle Route of South-to-North Water Diversion Project.