Research on flood forecasting in Tailan Watershed of Xinjiang based on PSO-TCN deep learning model

Accurate flood forecasting in advince is beneficial for planning flood measures in watersheds. In this study, a PSO-TCN (Particle Swarm Optimization-Temporal Convolutional Networks) flood forecasting model for the Tailan watershed in Xinjiang Uygur Autonomous Region was constructed by coupling the Particle Swarm Optimization (PSO) algorithm and the Temporal Convolutional Networks (TCN) algorithm. The model was tested using 50 historical flood events, based on observed rainfall-runoff data from 1960 to 2014 in the Tailing watershed. The results showed that under the same lead time conditions, the PSO-TCN model exhibited higher Nash efficiency coefficient, lower root mean square error, and lower relative peak error in flood forecasting. The PSO-TCN flood forecasting model demonstrated better applicability and robustness in the Tailan watershed. However, when the lead time exceeded 5 hours, the relative peak error of the PSO-TCN model still exceeded 20%. In the future, it is expected to integrate the mechanism of flood occurrence process to further improve the generalization ability of deep learning models in flood forecasting. The research results can provide reference for the calculation methods of flood forecasting in watersheds.