Logistic regression analysis to estimate contaminant sources in water distribution systems

Abstract

Accidental or intentional contamination in a water distribution system (WDS) has recently attracted attention due to the potential hazard to public health and the complexity of the contaminant characteristics. The accurate and rapid characterization of contaminant sources is necessary to successfully mitigate the threat in the event of contamination. The uncertainty surrounding the contaminants, sensor measurements and water consumption underscores the importance of a probabilistic description of possible contaminant sources. This paper proposes a rapid estimation methodology based on logistic regression (LR) analysis to estimate the likelihood of any given node as a potential source of contamination. Not only does this algorithm yield location-specific probability information, but it can also serve as a prescreening step for simulation–optimization methods by reducing the decision space and thus alleviating the computational burden. The applications of this approach to two example water networks show that it can efficiently rule out numerous nodes that do not yield contaminant concentrations to match the observations. This elimination process narrows down the search space of the potential contamination locations. The results also indicate that the proposed method efficiently yields a good estimation even when some noise is incorporated into the measurements and demand values at the consumption nodes.