483 - Feedback Control of Gravity Flow Systems Using Automatically Generated System Approximations

Control using distributed sensors, connected gates, valves, and pumps stands to transform management of urban stormwater systems by optimizing storage assets in real-time. However, the complexity and flashy dynamics of urban drainage systems limits the seamless application of available control approaches. Existing control algorithms, including machine learning, heuristic, and rule-based approaches, have limitations in computational expense, trust, coordination, and labor. Linear feedback stands out for its computational efficiency, trustworthiness, and built-in system-level coordination.
However, current methods for generating linear feedback controllers rely on existing calibrated software models and are not automated. Here we present an automated approach that requires only system connectivity and response data to generate linear feedback controllers. As it is invariant to time and spatial scales, the method has promising generalizability outside of urban drainage systems.
Using only response data and connectivity, we automatically construct a linear, time-invariant system of differential equations to approximate the hydraulic dynamics and enable linear feedback control. The controller is designed to maintain flows and depths below thresholds. The controller includes an observer to update system state estimates and a regulator to specify flows through valves. Though more complex strategies are also possible, tuning the controller’s performance is intuitive and customizable as it only requires specifying the maximum allowable depths and flows.
Performance is evaluated on a 4 km² semi-urbanized watershed under a 25-year, 6-hour storm, with equal-filling-degree control and uncontrolled responses as benchmarks. Only the linear feedback controller abides by all the flow and depth thresholds. This not only prevents flooding, but also improves stormwater treatment by significantly increasing hydraulic retention time.