Prediction of Residual Chlorine in Water Distribution Network Using Artificial Neural Network (ANN)

Protecting public health in rural water distribution networks requires reliable and safe disinfection. Although chlorination is the most widely used disinfection approach, determining the optimal chlorine dosage remains challenging because of environmental and distribution-related variations. This study presents an artificial neural network (ANN)-based predictive model for estimating residual chlorine in the Rui and Shingave Water Supply Scheme in Maharashtra, India. Field data, including variables such as pH, temperature, and distance from the Elevated Storage Reservoir (ESR), were collected from several nodes. A total of 250 samples were collected between 5 and 15 December 2024, covering winter-period variation. The min-max normalization method was used to standardize the dataset, and the NeuroSolutions v6 software was used for processing. The Levenberg-Marquardt algorithm was used to develop and train a Multilayer Perceptron (MLP) model, using 70% of the data for training and 30% for testing. Strong correlation coefficients, low MSE (mean squared error), and minimal MAE (mean absolute error) values across all phases showed the model’s high level of accuracy. Specifically, the model achieved R = 0.902, MAE = 7.18, and MSE = 114.25 on the test dataset. The results show that the ANN model can successfully learn the dynamics of chlorine degradation in the distribution network. With initiatives such as the Jal Jeevan Mission, this predictive method offers a practical tool for optimizing chlorine dosage in real time and strengthening water quality management in rural systems. Future integration with SCADA or IoT monitoring platforms may further enhance operational efficiency.