Pareto Optimality for Sensor Placements in a Water Distribution System

Abstract

As society looks to provide the most effective identification of possible intrusion events, issues of sensor placement in water distribution systems are drawing increased attention. A multiple objective optimization problem with two competitive objectives is formulated herein: (1) minimize time delay, and (2) maximize sensor detection redundancy. The two objectives are evaluated, based on a prebuilt database containing the array of potential intrusion events and detection information. Pareto fronts are developed to assess impacts of increasing numbers of sensors by nondominated genetic algorithm-II (NSGA-II). Further, Pareto front performance improvement of increasing numbers of sensors is quantified by average normalized Euclidean distance to identify the point of diminishing marginal return aiming to provide rationale for estimating the number of sensors needed for a water distribution system. A case study is conducted for the City of Guelph water distribution system. It is observed that increasing the numbers of sensors results in better performance on the Pareto front. The Pareto front improvement rate indicates five sensors as the point of diminishing marginal return, which provides a basis for determining the number of sensors needed for the Guelph water distribution system.