A decomposition-coordination approach for distributed real-time control of urban drainage systems

Real-time control has emerged as an effective method for increasing the operational efficacy of urban drainage systems (UDSs). Traditional centralized control faces challenges related to efficiency and reliability as UDSs become more complex, which underscores the need for distributed control in ensuring resilient system operation. However, existing distributed control approaches rely heavily on expert experience or historical operational data, limiting their adaptability and scalability. To address this issue, this study proposes a decomposition-coordination-based distributed model predictive control (DC-DMPC) approach that achieves global coordination without requiring prior knowledge or centralized optimization. The method decomposes the UDS into multiple independent subsystems and employs an alternating-direction optimization scheme to ensure flow coordination across the system. It was compared with the practically used rule-based strategy (RBC) and traditional centralized model predictive control (CMPC) in a 20 km² combined UDS in Jiujiang, China. The performance of the three strategies was evaluated under four rainfall events considering variation in spatial distribution under normal operation and emergency scenarios involving communication failure. Results show that: (a) Under normal conditions with homogeneous rainfall, DC-DMPC reduced average flood and CSO volumes by 82.5 % and 23.1 %, respectively, compared to RBC, and by 9.4 % and 12.3 % compared to CMPC. Under inhomogeneous rainfall, these improvements reached 80.0 % and 34.5 % over RBC, and 19.6 % and 25.8 % over CMPC, highlighting DC-DMPC's enhanced adaptability to uneven rainfall patterns. (b) Under communication failures, DC-DMPC exhibited a 9.4 % smaller resilience loss compared to CMPC, along with a 36.4 % reduction in CSO volume. These results indicate that DC-DMPC effectively enhanced the solving efficiency by reducing the dimensionality of the optimization problem, while maintaining robustness during emergencies. The proposed approach provides a prior-knowledge-free and scalable framework for improving the operational resilience of UDSs.