A Data-Driven Approach to Identification of Contamination Source and Sensor Placement in Water Distribution Systems

This paper presents a data-driven approach to identifying contamination sources and sensor placement in drinking water distribution systems. First, a dataset including multiple scenarios of contamination positions with varying demands is generated using EPANET simulation. Second, structured as multi-channel input tensors, a convolutional neural network (CNN) is built and trained with the multivariate time series of water flow and quality data. Third, optimal sensor placement for contamination detection in the water network is determined by maximizing the classification accuracy. A case study demonstrates that the trained model is able to identify contamination positions even under uncertainty. This study illustrates the potential of combining hydraulic modeling and deep learning to support real-time water quality monitoring and decision-making in water distribution systems.