Leakage localization in water distribution using data-driven models and sensitivity analysis

Water scarcity is becoming an increasing problem worldwide, and an issue compounding the problem is water leakage in the piping networks delivering potable/consumable water to end-users (Sensus, 2012). In this paper, we consider the problem of isolating leakages in water supply networks using reduced network models. Using a reduced order model of the network, the expected behaviour of the network can be estimated and then compared with actual measurements obtained from the network. The result of this comparison is a set of residuals which are used to isolate a leakage to a network node. The localization is based on a sensitivity matrix which captures the residuals’ sensitivities to leakages. As the reduced order model is adaptive based on measurements from the network, the reduced order model is plug-and-play commissionable. The calculation of the sensitivity matrix is based on an EPANET model of the network and is performed off-line.