Author(s): Kaige Chen; Kexuan Liu; Yan Long; Xiaohui Lei; Zhifeng Huang; Jiaolong Zhang; Wentao Wei

Linked Author(s): Xiaohui Lei

Keywords: Deep learning; Flood control; Model predictive control; Surrogate model

Abstract: Traditional urban flood control strategies are hampered by the computational intensity of physical models and the reactive logic of predefined rules. Model Predictive Control (MPC), a dynamic and intelligent approach, can optimize hydraulic performance but is underutilized due to this computational barrier. This study introduces an integrated framework that incorporates a deep learning surrogate into the MPC architecture, effectively bridging the gap between high-fidelity simulation and real-time demands. A case study conducted in the Baishi Chong catchment in Zhongshan City, China, demonstrates the effectiveness of this framework. The results indicate that MPC outperforms the traditional Rule-Based Control (RBC) strategy, particularly under frequent storm conditions. Notably, MPC not only provides superior hydraulic benefits but also significantly enhances ecological conditions. At a critical monitoring point within the main river channel, it more than tripled the duration for ecological water level maintenance while it eliminated the water level alert periods prevalent under the RBC approach. Such findings suggest that integrating a deep learning-driven MPC is a highly effective alternative to reactive scheduling for flood management, which significantly enhances the operational performance of existing hydraulic infrastructures without substantial new construction.

DOI:

Year: 2026