Localization of Multiple Existing Leakages in Water Distribution Networks by a Graph-Based Genetic Algorithm

Managing non-revenue water in water distribution networks (WDNs) caused by leakages is of greatest interest. In particular, accurately identifying the spatial locations of multiple leakages within the WDNs for the on-site search is both a challenging and a computationally expensive task, especially when using genetic algorithms (GA). To overcome this challenge, this work proposes a graph-based GA for the localization of multiple leakages. Unlike classical GAs, the presented approach integrates topological information during reproduction, accounting for the spatial relationships between potential leakage points. The graph-based GA was evaluated using 100 leakage scenarios with varying characteristics, achieving a median distance of 80 m between the predicted and actual leakage point under perfect conditions. This demonstrates that the graph-based is a highly accurate and computationally efficient alternative to classical GAs, though its accuracy is influenced by uncertainties inherent in model-based leakage localization.