Published: Građevinar 78 (2026) 9
Paper type: Original scientific paper
Download article (Croatian): PDF
Download article (English): PDF

Leak location and magnitude estimation in water distribution networks using an optimised two-stage leak localisation model

Water loss from urban water distribution networks (WDNs) challenges sustainable water management and utility operations. Early leak detection and accurate localisation are crucial for reducing non-revenue water and maintaining network efficiency. This study proposes a machine learning-based framework integrating EPANET-generated static hydraulic simulations with an artificial neural network (ANN) optimised using particle swarm optimisation (PSO) for effective feature extraction. Pressure data under various leak and no-leak scenarios were simulated, and sensitive sensor nodes were strategically selected to enhance accuracy. By bridging simulation and reality, this method successfully integrates hydraulic modelling with data-driven learning, demonstrating the hybrid framework’s strong potential for scalable and accurate leak detection, localisation, and magnitude estimation in WDNs.

LICENCE:

This paper is licensed under a Creative Commons Attribution 4.0 International License.