Operational efficiency improvement in a water supply network: Machine learning-enhanced leakage identification and water resource conservation

Abstract

Pipeline ruptures in water supply networks can induce significant water loss and may pose risks of water quality deterioration, including potential contamination by pathogens and pollutants. This issue can be addressed by predicting the location of leakage points in the pipeline network and controlling the leakage. Here we designed a hydraulic model for leakage localization using a genetic algorithm-backpropagation neural network, to predict the leakage points in the water supply system of an exposition area consuming 117,211 m³ of water per day in Eastern China. Then, using the model results, pressure-regulating valves were installed in areas with lower network safety. Results show that the error in predicting the leakage points localization ranged from 14.48 m to 121.69 m. The installation of pressure-regulating valves, reduced the average water pressure from 33.54 m to 32.64 m (2.7 %) and, in turn, decreased the simulated background leakage by 9684 m³ of water per day. Compared to traditional acoustic-based methods, the proposed machine learning approach enables more accurate leak localization by leveraging pressure variation features.