Source Contamination Detection and Identification of Affected Spots in the Intermittent Water Supply System Using Mixed Tool-Based and Data-Driven Machine Learning Techniques

It has always been difficult to model the real-life intermittent water distribution system. This can be achieved by adopting operational setups of the modeling tool. The Gajrawadi water supply network of Vadodara City, India, serves as the case study. In EPANET, a hydraulic model is developed using pressure-driven analysis and rule-based controls to address intermittent water supply. A chlorine decay kinematics is used for water quality simulation. Simulation results are generated by capturing scenarios of leakage and quality degradation using a leak reservoir approach and local bulk reaction coefficient adjustments. These features, along with operational strategies, closely relate to the real-world contamination intrusion event of sewage-mixed accumulated water in soil cavities, which enters through a localized leakage spot. Datasets generated from such scenarios are used to obtain residual chlorine-based conditional probabilities. This also ensures the absence or presence of contamination in the system. The classification datasets are further processed using the support vector machine and an artificial neural network. These models identify the probable locations of Zone 4, where the contaminated water has reached, and detect source contamination with 83% and 92.39% model accuracy, even under system leakage (0.061% to 0.378%).