Improving Analytics in Urban Water Management: A Spectral Clustering-based Approach for Leakage Localization

Worldwide growing water demand has been forcing utilities to successfully manage their costs. Contemporarily, within an era of tight budgets in most economic and social sectors, it affects also Water Distribution Networks (WDN). So, an efficient urban water management is needed to get a balance between consumer satisfaction and infrastructural assets inherent to WDN. Particular case is referred to pipe networks which suffer for frequent leaks, failures and service disruptions. The ensuing costs due to inspection, repair and replacement, are a significant part of operational expenses and give rise to difficult decision making. Recently, the goal regarding the improvement of the traditional leakage management process through the development of analytical leakage localization tools has been brought to the forefront leading to the proposal of several approaches. The basis of all methods relies on the fact that leaks can be detected correlating changes in flow to the output of a simulation model whose parameters are related to both location and severity of the leak.This paper, starting from a previous work of the authors, shows how the critical phases of leak localization can be accomplished through a combination of hydraulic simulation and clustering. The research deals with the benefits provided by Spectral Clustering which is usually adopted for network analysis tasks (e.g., community or sub-network discovery). A transformation from a data points dataset, consisting of leakage scenarios simulated through a hydraulic simulation model, to a similarity graph is presented. Spectral Clustering is then applied on the similarity graph and results are compared with those provided by traditional clustering techniques on the original data points dataset. The proposed spectral approach proved to be more effective with respect to traditional clustering, having a better performance to analytically localize leaks in a water distribution network and, consequently, reducing costs for intervention, inspection and rehabilitation.