Leakage Area Detection in Water Distribution Networks: a Two-Step Algorithm Using Limited Data

This paper presents a two-step algorithm for leak detection in Water Distribution Networks (WDNs). In the first step, a minimization process based on a derivative-free optimizer reduces the difference between simulated and measured data at the respective pressure/flow sensors installed within the WDN. Leakage coefficients serve as decision variables, each unequivocally associated with a specific location within the network. In the second step, a filtering-clustering-ranking algorithm eliminates nodes where the leaked volume is considered negligible, identifies leakage clusters, and generates a priority list of nodes for further inspection using leak isolation or pinpointing techniques. The proposed algorithm falls within the category of leakage awareness models; therefore, it is not designed to perfectly localize leaks but rather to narrow down the leakage area, facilitating subsequent on-field localization efforts. Tests conducted on a realistic WDN demonstrated promising performance, as the algorithm successfully restricted the leakage area—sometimes even pinpointing the exact location—using a limited number of pressure/flow sensors, despite varying leak magnitudes and locations across the network. Preliminary tests also indicated that the algorithm performs well in scenarios involving multiple leaks. Overall, the proposed model was capable of including the leaking node within a priority list 96% of the time, considering all the parameter settings used and all the leakage scenarios considered.