Machine Learning-based Optimal Control of Sewer Systems to Minimize Sewer Overflows

A combined sewer system (CSS) collects rainwater runoff, domestic sewage, and industrial wastewater in the same pipe. The volume of wastewater can sometimes exceed the system capacity during heavy rainfall events. When this occurs, untreated stormwater and wastewater discharge directly to nearby streams, rivers, and other water bodies. The overflows threaten public health and the environment, contributing to drinking water contamination and other concerns. Minimizing sewer overflows require an optimization solver to determine the optimal sequence of gate position (decision variables) at control units. Conventional optimization strategies use classical optimization algorithms, such as Genetic algorithms and Pattern search. However, these conventional frameworks are time-consuming and unfeasible for real-time optimal control. This paper presents a new optimal decision tool using a two-stage deep learning strategy. In the first stage, we develop a surrogate time series forecasting deep learning model to predict water stage, overflow location, and overflow volume for a fixed position of the controlled gates. The training dataset is obtained by SWMM model. In the second stage, we use reinforcement learning to create a sequence of input “policy” that is fed into the first stage model. The reward function is constructed based on the states from predictive deep learning models (overflow rates, water depths). The process minimizes the reward function to obtain the optimal flow release sequence at each control gate. The combined sewer system (CSS) of the Puritan-Fenkell 7-mile Facility in Detroit, Michigan, is used as the case study.