Contamination source inference in water distribution networks

TL;DR

This paper presents a method to identify contamination sources in water networks by modeling contamination spread, using sensor data, and solving an optimization problem with mixed integer linear programming, tested on real and synthetic networks.

Contribution

It introduces a novel optimization-based approach for contamination source inference in water networks using simplified dynamics and sensor data.

Findings

Effective source localization in simulated networks

Successful application to Modena city water network

Optimization approach outperforms baseline methods

Abstract

We study the inference of the origin and the pattern of contamination in water distribution networks. We assume a simplified model for the dyanmics of the contamination spread inside a water distribution network, and assume that at some random location a sensor detects the presence of contaminants. We transform the source location problem into an optimization problem by considering discrete times and a binary contaminated/not contaminated state for the nodes of the network. The resulting problem is solved by Mixed Integer Linear Programming. We test our results on random networks as well as in the Modena city network.