(644g) An Application for Real-Time Response to Contamination Events in a Large-Scale Public Water Network

The accidental or in­­tentional injection of chemical or biological contaminants into the drinking water network poses a serious security threat to the people living in modern cities. In case of a contamination event, ensuring the supply of safe, clean drinking water requires efficient detection, source determination, and response planning. Mathematical programming provides an invaluable tool for optimal design of early warning detection systems, and optimal real-time response and mitigation. In this presentation, we discuss several mixed-integer linear programming formulations that form part of a real-time security application.

Working directly with the Public Utilities Board in Singapore, we have developed the software package Dinesti, which interfaces with utility SCADA systems to provide real-time response optimization for water quality events. The CANARY system, developed at Sandia National Laboratories, uses standard water quality data coupled with statistical methods to identify the onset of abnormal water quality events. Dinesti utilizes input from continuous monitoring of water quality provided by CANARY and data from “grab” samples taken at discrete times and locations within the network. Dinesti’s real-time response capabilities include numerical source identification and optimal location determination for manual sampling. Source identification requires the solution of a large-scale mixed-integer linear programming problem with an embedded water quality model for the network. Given the large network size, efficient solution of the source identification problem is only possible because of a novel reduction strategy for the linear water quality model. Given limited sensor information, it is likely that the algorithm will identify a large family of potential source locations. As part of the real-time response application, we also include a strategy to help narrow down the true source node by optimally selecting manual sampling locations. These locations are determined by solving a mixed-integer linear programming formulation that maximizes the distinguishability of potential events (a maximum coverage problem). The utility can mobilize teams to gather manual samples at these locations for analysis. This additional data is included in the next cycle, where source identification is performed again.

In this presentation, we will discuss the Dinesti application, and we will present the optimization formulations used to perform source identification and optimal sampling location determination. Special attention is given to development of mathematical and computational techniques that aid in achieving fast real-time performance