Fast Detection and Localization of Multiple Leaks in Water Distribution Network Jointly Driven by Simulation and Machine Learning

The leakage control in water distribution networks (WDNs) is of high concern in the water supply industry. One direct and effective way to reduce leakage is to adopt leakage detection and localization methods to guide water utilities to repair broken pipes in time. In order to achieve higher accuracy in the leakage detection process in WDN with multiple leaks, a novel multiple leak detection and localization framework (MLDLF) based on existing pressure and flow measurements is proposed. The MLDLF decomposed the problem into three substages: model calibration, leakage identification, and leakage localization. After using the calibrated hydraulic model to predict pressure values and estimate overall leakage flow in each area in the first stage, the data-driven methods, STLK, including the seasonal and trend decomposition using loess (STL decomposition) and the $k$-means clustering method, were performed in the identification stage to distinguish different leakage scenarios so as to determine the occurrence time of every leakage event. Finally, combined with the stepwise model–based fault diagnosis method, leakages were located gradually with high computational efficiency. A case study of applying MLDLF to the WDN of L-Town showed that 56.52% of the leakage events were successfully identified and located with the economic score reaching €264,873, indicating the robustness and good applicability of MLDLF in identifying and localizing all types of leaks under multiple leakage scenarios.