PIML-Driven Spatial Optimization and Active Telemetry Control of Integrated Grey-Green Systems to Minimize Modernization Capital Investment in US Combined Sewer Systems
Year: 2026
Type: preprint
Abstract: Combined sewer overflow (CSO) mitigation requires reliable hydraulic prediction and cost-effective infrastructure planning. This study develops a physics-informed machine learning (PIML) framework integrating surrogate modelling, real-time control (RTC), and spatial green-grey infrastructure optimization to minimize capital investment in combined sewer systems. The framework was evaluated using the 11.7-ha EPA SWMM Example 8 benchmark, NOAA Atlas 14 design storms, and 20 years of hourly rainfall data from Philadelphia. Data-driven and physics-informed surrogates were compared through one-step prediction, closed-loop rollout, and extreme-storm evaluation. Rule-based control (RBC) and model predictive control (MPC) were assessed across 15 design storms, while NSGA-II optimized infrastructure configurations and capital costs. PIML-soft achieved a CSO-flow rollout NSE of 0.87 under test conditions and 0.93 for withheld 25-100-year storms, outperforming the data-only MLP in extended prediction. RBC reduced CSO volume by 5–29% relative to passive operation and outperformed the tested MPC configurations. The selected GI-led design, comprising 38 bioretention units and 6,000 gallons of residual storage, reduced modal capital cost from $0.87 million to $0.49 million compared with grey-only storage. These findings demonstrate the potential of integrated PIML and spatial optimization for benchmark-specific, cost-effective CSO management.
Source: Research Square
Author: Naim Sheikh
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7236502584
