Data-driven Leak Localization in Water Distribution Networks via Dictionary Learning and Graph-based Interpolation

In this paper, we propose a data-driven leak localization method for water distribution networks (WDNs) which combines two complementary approaches: graph-based interpolation and dictionary classification. The former estimates the complete WDN hydraulic state (i.e., hydraulic heads) from real measurements at certain nodes and the network graph. Then, we append to the actual measurements a subset of relevant estimated states to feed and train the dictionary learning scheme. Thus, the meshing of these two methods is explored, and several promising performance results are attained, even deriving different mechanisms to increase the resilience to classical issues (e.g., dimensionality, interpolation errors, etc.). The approach is validated using the L-TOWN benchmark proposed in the BattLeDIM2020 competition.