Supporting Integrated Operation of Sewer Networks and Wastewater Treatment Plants with a Convolutional Neural Network-Long Short-Term Memory-Attention Model

Ensuring stable water quality at sewer network end points is important for improving the efficiency of the integrated operation of sewer networks and wastewater treatment plants (WWTPs). Accurate prediction of end point water quality can guide wastewater allocation among WWTPs, enhancing system operational efficiency. In this study, a convolutional neural network-long short-term memory-attention (CNN-LSTM-Attention) model was developed to predict sewer network water quality, achieving a prediction accuracy of 80.2%. Analyzing the contribution of each submodel, the results showed that LSTM captured temporal dependencies, while CNN enhanced the extraction of key features and the attention mechanism dynamically adjusted feature weights to prevent overfitting. Analysis of different input feature combinations revealed that, with low-quality data, using single water quality features could serve as an effective alternative. By optimizing the input sequence length and data hierarchy, the model achieved its best performance with a 48 h input sequence and optimal data hierarchy, reaching a prediction accuracy of 90.7%. In conclusion, the CNN-LSTM-Attention model, driven upstream data input, provided reliable predictions of water quality at sewer network end points. These predictions enhanced decision-making for influent wastewater allocation across multiple WWTPs, thereby mitigating the impact of sewer network fluctuations. This work contributes to the stable and integrated operation of municipal wastewater treatment and management.