<p>Accurate and fast prediction of water quality in urban river networks is essential for sustainable river ecosystem management and rapid emergency response decision to water pollution. Artificial intelligence methods have become a promising preferred approach. However, in the face of high human interference and dramatic natural climate change, even machine learning models such as artificial neural networks can face some dilemmas, leading to a decline in the effectiveness of predictions. To overcome the limitations of empirical hyperparameter selection in LSTM-based water quality prediction, this study introduces a novel hybrid model, SSA-LSTM, that integrates the sparrow search algorithm (SSA) with long short-term memory (LSTM). This model is applied to predict chemical oxygen demand (COD_Cr), ammonia nitrogen (NH₃-N), and total phosphorus (TP) in the Haihe River network in Tianjin, China. A key novelty of this study lies in incorporating the water quality conditions under 26 simulated management scenarios into the training and validation data of the model, exposing the model to extensive human interference. The performances of Multi-layer perceptron (MLP), LSTM and SSA-LSTM were verified and compared. The results show that the average R² of SSA-LSTM for the three water quality parameters of the upper, middle and lower reaches of secondary rivers is 0.81, which has a good prediction effect. This study provides a fast and accurate water quality prediction method for complex urban river networks.</p>