Data-driven modelling for characterising the behaviour of urban sewer networks

Background Urban sewer networks are increasingly exposed to hydraulic stress due to stormwater inflows and infiltration, usually known as parasitic flows. They cause extra loads on the sewer network, especially during rainfall, leading to higher treatment costs, potential overflows and operational issues.Methods This study employs a symbolic machine learning approach to characterise urban sanitary sewers in relation to the so-called “parasitic flows”, which are simply defined as unwanted flows introduced into the sewer system. Specifically, we used the Evolutionary Polynomial Regression (EPR) technique, which combines genetic programming with numerical and symbolic regression, to develop simple, easily interpretable mathematical expressions.Results The framework detects operational discontinuities, providing evidence for informed management decisions and demonstrating the tool’s effectiveness for diagnostic benchmarking, while also establishing a methodological basis for identifying the operational variables required to define a systematic monitoring framework for sanitary sewer networks.Conclusions This approach establishes a foundation for the proactive management of sewer-treatment systems and supports asset management planning in line with the objectives of the new Directive (EU) 2024/3019 on urban wastewater treatment.