AI-Driven Leak Detection in Water Distribution Systems: A Machine Learning Approach

This research investigates the application of machine learning (ML) techniques for leakage detection in water distribution systems (WDS), addressing the significant economic and infrastructural challenges posed by water losses. Utilizing the EPANET simulation model, we generated water pressure data and applied it to a real-world WDS case study in Ontario, Canada. The study begins by analysing flow and pressure data obtained from EPANET, which were subsequently used to evaluate the effectiveness of popular ML models in managing leakage. This approach allows for a comprehensive assessment of how well different ML models including Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Learning Machine (ELM) can detect leakages within a sample WDS. Models’ performance was evaluated using key statistical metrics. Among the models evaluated, the ELM demonstrated superior performance in both training and testing phases, as indicated by scatter plot visualizations and key performance metrics. Specifically, the ELM model achieved high correlation coefficients (R = 0.915 and 0.908) and low error values (RMSE = 0.201 and 0.223; MAE = 0.117 and 0.115) for training and testing, respectively. This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).