Pollution-Based and Multi-Layer Model Predictive Control of Urban Drainage Systems

Urban Drainage Systems (UDS), particularly Combined Sewer Overflows (CSO), face challenges in managing overflows during heavy rainfall, often leading to a negative environmental impact. Traditional CSO mitigation methods are costly, prompting interest in software-based strategies such as Model Predictive Control (MPC). This research focuses on enhancing Volume-based MPC through two complementary frameworks: (1) a pollution-based framework, and (2) an optimisation-based RTC framework, integrating pollutant concentrations into the control process to prioritise treatment of highly contaminated flows. To address practical limitations, this work introduces a Pollutionweighted MPC (PWMPC) for fast linear optimisation, a Non-linear Pollution-based MPC (NPMPC) for best load reduction performance, and a multi-layer MPC framework to ensure feasible solutions under real-time constraints. Stochastic MPC variants are also explored to account for uncertainty in system dynamics. The results show that the multi-layer MPC and its stochastic counterpart improve reliability, reducing infeasibility instances in NPMPC. Although NPMPC offers the best performance in low- to mid-uncertainty conditions, stochastic methods are more robust under high uncertainties. Furthermore, PWMPC is identified as the most computationally efficient, but depends heavily on accurate concentration data. Finally, Volume-based MPC remains useful when concentration data is unavailable and as an ultimate fallback. The study also presents a methodology for weight selection in all types of MPCs, which improves tuning and performance. Ultimately, NPMPC was found to deliver nearoptimal pollutant reduction, making it an attractive strategy to reduce overflow load. Other MPC implementations, while approaching optimal solutions, do not exceed the performance of NPMPC in most scenarios, establishing it as the most effective strategy. However, for practical implementations, both PWMPC and SPWMPC are arguably the most cost-effective method due to their powerful performance on reducing load with a low computational cost. Finally, the Multi-layer MPC improved the performance by enabling a better use of NPMPC or SNPMPC methods in complex systems.