Contamination Source Detection in Water Distribution Networks using Belief Propagation

TL;DR

This paper introduces a Bayesian belief propagation method to identify the most probable contamination source in water networks based on sensor data, using a simplified binary model validated on realistic simulations.

Contribution

It presents a novel Bayesian approach employing belief propagation for contamination source detection, integrating a simplified binary dynamics model with realistic network simulations.

Findings

Effective source identification in simulated water networks

Belief propagation accurately estimates source probabilities

Simplified model captures key contamination information

Abstract

We present a Bayesian approach for the Contamination Source Detection problem in Water Distribution Networks. Given an observation of contaminants in one or more nodes in the network, we try to give probable explanation for it assuming that contamination is a rare event. We introduce extra variables to characterize the place and pattern of the first contamination event. Then we write down the posterior distribution for these extra variables given the observation obtained by the sensors. Our method relies on Belief Propagation for the evaluation of the marginals of this posterior distribution and the determination of the most likely origin. The method is implemented on a simplified binary forward-in-time dynamics. Simulations on data coming from the realistic simulation software EPANET on two networks show that the simplified model is nevertheless flexible enough to capture crucial information about contaminant sources.