Machine-learning-assisted prediction and optimized kinetic modelling of residual chlorine decay for enhanced water quality management

Abstract

The quality of water changes from source to tap, presenting challenges in maintaining consistent water quality across the system. Predicting water quality in distribution systems, including disinfectant residual loss and by-product formation, has been the subject of research since the early 1990s. Although numerous models have been proposed to predict residual chlorine decay, disputes exist among researchers and experts over the superiority of certain models. Accordingly, this study modified the existing process-based bulk decay models by replacing the initial Total Residual Chlorine (TRC) concentration parameter with TRC demand, leading to an improvement in the models' performance. The modification resulted in a 38.03%, 28.02%, 23.11%, and 33.29% average improvement in Mean Squared Error (MSE) values for the First Order Model (FOM), Parallel First Order Model (PFOM), Second Order Model (SOM), and Parallel Second Order Model (PSOM), respectively. The study also introduced an online predictive method based on a Machine Learning (ML) algorithm that predicts the first-order TRC bulk decay rate by using water quality parameters as inputs. A Gaussian Process Regression (GPR) model was used to predict the kinetic parameters in FOM, which accurately predicted the test sets for most of the cases. In addition, a new methodology was proposed in this study for predicting TRC in water distribution systems that incorporates the variability of source natural organic matter, operational actions, and water demands. This method seeks to develop high-fidelity and robust water quality predictions that provide operational decision support for optimized distribution system management. In conclusion, this study emphasizes the importance of understanding water quality changes from source to tap and the challenges of maintaining consistent water quality across the system. The study suggests modifying existing models and introducing a novel methodology for predicting residual chlorine in water distribution systems that can improve water quality management and, ultimately, better public health outcomes.

Keywords: Bulk decay; Chlorine decay; Distribution system; Machine learning; Modelling; Water quality.