Time-series prediction of water quality at multiple points in water sources

Abstract

Effective monitoring and prediction of water quality is essential for the management and protection of aquatic ecosystems. Although great progress has been made, there still exist numerous challenging cases, such as the lack of consideration for global changes in water quality and comprehensive factors affecting it. To address this limitation, a Multiple-View Online Long Short-Term Memory (MV-Online-LSTM) model was proposed, which integrated multi-view learning with online sequential adaptation. Each monitoring points were treated as a separate view processed by dedicated LSTM sub-networks, which were fused to capture spatial dependencies and temporal dynamics. An online learning strategy enabled real-time model updates, enhancing adaptability to environmental changes. Experiments demonstrated that MV-Online-LSTM achieved R² values consistently above 0.96 across six key water quality indicators, significantly outperforming baseline models. These findings underscore the effectiveness of the proposed model in dynamic, multivariate water quality forecasting, offering a practical solution for real-time environmental monitoring applications.