Research on source identification of river sudden water pollution based on PINN

With the continuous industrialization, sudden water pollution events are characterized by increasing abruptness and complex dispersion patterns. Such events often cause cascading damage to aquatic ecosystems and result in significant economic losses, making fast and accurate pollutant source identification and scientific emergency response critically important. Traditional source identification methods frequently encounter challenges such as heavy computational burdens and low efficiency. To address these issues, this study introduces an artificial intelligence-integrated model. By employing deep learning to extract the spatiotemporal variation characteristics of river channels and coupling them with pollutant transport and diffusion equations, a fast identification method for sudden water pollution in rivers channels based on the Physics-Informed Neural Network (PINN) is proposed. The specific contents are as follows: (1) building a PINN-based model for sudden water pollution source identification in river channels and estimating the detailed information of pollution sources; (2) improving the PINN algorithm to further enhance identification accuracy; (3) applying the proposed method to a typical river channel case, with results demonstrating that the method can effectively identify the emission intensity, location, and time of sudden water pollution sources. The findings indicate that the PINN-based identification method maintains good accuracy and robustness even under conditions of sparse monitoring data. This study provides a new perspective for solving the inverse problem of pollution source identification in river systems.