A comparison of machine learning classifiers for leak detection and isolation in urban networks

Abstract

[EN] Leak detection and isolation (LDI) is a problem of interest for water management companies and their technical staff. Main reasons for this are that early detection of leakages can reduce dramatically (1) water losses in urban networks and (2) the environmental burden due to wasted energy used in the system supply [1]. Water leakage can become a very complex problem, due to the lack of information about the water system and because a leak might not be easily detected on-sight. Therefore, any diagnostic tool that could help in such task are valuable for engineers and managers. Soft computing tools have shown to be valuable tools for researchers in different fields. Supervised machine learning techniques for example, have been used with success in complex problems, for binary and multi class classification. This is useful in order to detect different faulty scenarios in complex systems using for example, on-line data from SCADA systems. In this paper, we provide an analysis on some soft computing techniques used for LDI in urban networks. This with the aim of identifying strengths and drawbacks among different machine learning techniques for this task in real-time acquisition scenarios.