Real-time Contaminant Source Characterization in Water Distribution Systems

Abstract

Accidental/intentional contamination continues to be a major concern for the security management in water distribution systems. Once a contaminant has been initially detected, an effective algorithm is required to recover the characteristics of the contaminant’s source based on dynamically varying streams of sensor observations. This dissertation focuses on the development and demonstration of a new algorithm to characterize a contaminant source quickly, accurately, and robustly. An evolutionary algorithm (EA)-based adaptive dynamic optimization technique (ADOPT) is proposed, potentially providing a real-time response. In addition to offering adaptive capacity in a dynamic environment, this algorithm is able to assess the degree of non-uniqueness in the solution through multi-population scheme. This approach, however, requires a large number of time-consuming simulation runs to evaluate possible solutions, and it may be difficult to converge on the best solution or a set of alternative solutions within a reasonable computational time. For this reason, it is desirable to appropriately reduce the decision space over which the optimization procedure must search to reduce the computational burden and to produce faster convergence. A logistic regression-based prescreening technique is investigated in order to reduce the decision space by estimating the probability of a node being a contaminant source location. When a small set of potential source nodes are identified, applying the local search procedure to this set of locations is computationally efficient and potentially good at identifying the best solution. The EA-based ADOPT is then integrated with a logistic regression analysis and a local improvement method to expedite the convergence and to solve the problem potentially faster. The effectiveness of the proposed methods is demonstrated for contamination source identification problems in two illustrative water distribution networks.