Physics-Informed Machine Learning for Co-Optimized Green Infrastructure and Real-Time Control of Combined Sewer Systems: Capital Cost and Traffic Disruption Assessment
Year: 2026
Type: preprint
Abstract: Combined sewer systems can discharge untreated combined sewer overflows (CSOs) during intense rainfall, while conventional storage tunnels require substantial capital investment and may disrupt urban traffic during construction. This study proposes an integrated modernization framework combining Non-dominated Sorting Genetic Algorithm II (NSGA-II)-based placement of roadside bioretention cells with real-time gate and pump control driven by a physics-informed machine-learning (PIML) surrogate. The surrogate incorporates the one-dimensional Saint-Venant mass- and momentum-conservation equations into its training objective. The framework was evaluated using an EPA SWMM 5.2-based benchmark catchment under 1-, 2-, 5-, and 10-year synthetic design storms. Against the supplied SWMM benchmark outputs, the PIML surrogate achieved a Nash–Sutcliffe efficiency (NSE) of 0.952 and a reported inference latency of 0.32 ms, compared with a mean SWMM runtime of 14.5 min. The integrated scenario reduced peak runoff from 48.2 to 19.6 m³/s (59.3%) and increased reported storage utilization from 42.0% to 94.5%. Simulated CSO discharge was reduced to zero for the 1-, 2-, and 5-year storms and from 210,000 to 12,400 m³ for the 10-year storm (94.1%). Road-waterlogging duration decreased from 4.2 to 0.3 h (92.9%). Using the cost components stated in the manuscript, baseline capital cost is $250 million and proposed capital cost is $20 million, corresponding to an estimated 92.0% reduction; regulatory penalties are excluded from capital expenditure. These are model-based estimates and require reproducible model files, cost documentation, and field validation before practical deployment.
Source: Preprints.org
Authors: Naim Sheikh, Lamia Islam Lima
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7226010599
