Model predictive control based on artificial intelligence and EPA-SWMM model to reduce CSOs impacts in sewer systems

Abstract

Urbanization and an increase in precipitation intensities due to climate change, in addition to limited urban drainage systems (UDS) capacity, are the main causes of combined sewer overflows (CSOs) that cause serious water pollution problems in many cities around the world. Model predictive control (MPC) systems offer a new approach to mitigate the impact of CSOs by generating optimal temporally and spatially varied dynamic control strategies of sewer system actuators. This paper presents a novel MPC based on neural networks for predicting flows, a stormwater management model (SWMM) for flow conveyance, and a genetic algorithm for optimizing the operation of sewer systems and defining the best control strategies. The proposed model was tested on the sewer system of the city of Casablanca in Morocco. The results have shown the efficiency of the developed MPC to reduce CSOs while considering short optimization time thanks to parallel computing.

A review of pollution-based real-time modelling and control for sewage systems

“…Research studying the uncertainty in and how to handle it in RTC systems is limited [55] . Robustness is also important for keeping stability, but research in this field is not existing for P-RTC methods; There are some studies in stochastic volume-based RTC ( [82] and [33] ) and drinking water networks ( [24] and [18] ), however using this framework within pollution-based RTC is presently missing and this could be an interesting research field; Data-driven models for prediction in SUDS are a well-known topic in literature and the application of these models in volume-based RTC can be observed in [5] , [23] and [45] . However, the research of data-driven methods for water quality prediction used in P-RTC is still in its infancy, thus being an appealing research topic for future research; Further developments on reinforcement learning RTC that is based solely on the research done by [10] , such as different reward shapes, applying a multi-objective RL P-RTC to deal with both quantity and quality targets or study RL optimisation for different pollutant types; More studies in real applications of pollution-based RTC.…”

“…Data-driven models for prediction in SUDS are a well-known topic in literature and the application of these models in volume-based RTC can be observed in [5] , [23] and [45] . However, the research of data-driven methods for water quality prediction used in P-RTC is still in its infancy, thus being an appealing research topic for future research;…”

Global Paradigm Shifts in Urban Stormwater Management Optimization: A Bibliometric Analysis

“…Yang et al [76] proposed a machine learning method to learn the hydrological response of sustainable urban drainage systems. El Ghazouli et al [77] proposed a novel model based on neural networks for predicting flows, a stormwater management model (SWMM) for water conveyance, and a genetic algorithm for optimizing sewer system operations and defining optimal control strategies. These can effectively reduce combined sewer overflows.…”

Combined Optimization of LID Patches and the Gray Drainage System to Control Wet Weather Discharge Pollution

“…SWMM can simulate the storm transport process in urban drainage systems , in which various factors from runoff generation and confluence to drain-flow transportation are taken into account. ,, Using the SWMM in combination with ESRI ArcGIS software, the boundaries of subcatchments were divided based on the land use, runoff path, and the layout of storm drainages in the study area, and 58 subcatchments were formed (ZMJ1-ZMJ58, Figure S1). According to the principle of proximity, the runoff of each subcatchment flowed into the nearest node (inspection well) in storm drainages.…”