Robust Prediction of Residual Chlorine Decay in a Drinking Water Distribution System Integrating Water Quality Sensing and Predictive Tools

This research paper presents a novel approach for predicting residual chlorine in water distribution systems, a critical component for ensuring safe drinking water quality. This new method innovatively incorporates the variability of source Natural Organic Matter (NOM), operational actions, and water demands. It combines water quality sensing with data-integrated hydraulic and chemical reaction models, aiming to offer high-fidelity and robust water quality predictions and, thus, provide operational decision support for optimized distribution system management. Until now, the absence of real-time data and appropriate algorithms to quantify and analyze source quality characteristics has inhibited the development of a robust water quality predictor. This research fills this gap and enables direct predictions of Total Residual Chlorine (TRC) instead of using water age as a surrogate. By integrating kinetics modeling approaches and tracking NOM variations in the network as well as implementing the predictive tools, it is now possible to anticipate TRC decay rates and manage water quality more effectively. Three distinct models were evaluated in this study: two pure black-box models leveraging Artificial Neural Networks (ANNs) and the Gaussian Process Regression (GPR) algorithm and a hybrid model combining a first-order kinetic model with a GPR algorithm. The results showed significant improvement in MSE values, up to 90%, in prediction accuracy for TRC in the distribution network compared to traditional kinetic models.