Using symbolic machine-learning as a metamodel for chlorine decay in a water supply network

The timely assessment of disinfection in water distribution networks is fundamental to guarantee the compliance of quality standards, such as chlorine residuals. This work presents the application of a symbolic machine-learning strategy named Evolutionary Polynomial Regression (EPR) as a metamodel to estimate chlorine concentrations. This model is presented as an alternative to the inherent uncertainty in the calibration of physically based models when only sparse measurements are available. The methods are demonstrated in a real water supply network, used to generate synthetic chlorine decay data following first-order kinetics for training EPR. The resulting models have a high estimation accuracy, and they show that simpler analytical formula may be used for mostly branched networks if the changes in boundary conditions are considered. This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).