Accurate and Optimal Sensor Placement for Source Identification of Water Distribution Networks

Abstract

The problem of sensor placement for early-warning detection systems is a topical issue for industry and utilities who want to equip their networks with such technology. Solving the problem entails finding the best sensor locations that optimize a criterion such as detection rate or time to detection. Few methods exist concerning sensor placement that optimize the result of a source identification method. This paper fills the gap by coupling an adjoint source identification method and a Monte Carlo sensor placement algorithm. The first one is treated through the use of a backtracking algorithm. It uses binary responses at sensors to calculate the ranked list of potential contamination location nodes and contamination times. A criterion is then defined based on the source identification accuracy and specificity. Finally, two optimizing methods that maximize this criterion are proposed: a greedy algorithm and a local search algorithm, which are both coupled with a Monte Carlo method to give the locations of sensors that are the best suited for allocating the source of a contamination. These methods are tested on a 2,500-node network to evaluate their efficiency.