(778c) Optimal Design and Operation of Real-Time Response Systems for Water Security

Drinking water distribution systems are vulnerable to accidental or intentional contamination. Contaminants can ingress into the system through pipe breakages, water sources, or outlet (e.g. fire hydrant, household faucets) backflows. One proposed approach is the installation of an contaminant warning detection system – a sparse layout of sensors to detect the presence of contaminant. Effective control and cleanup of the system is also essential to reduce population exposure to the contaminant and to return the system to an operational state. Modern mixed-integer programming strategies play a role in the design of protection systems and optimal execution of a response strategy. In conjunction with researchers at Sandia National Laboratories and the Environmental Protection Agency, we have developed several advanced problem formulations to protect the public against potential contamination events. These tools have been integrated into the EPA’s Water Security Toolkit, an open-source software package being developed for use by water utilities. In this presentation, we will show three components of the Water Security Toolkit.

First, we developed a novel water quality simulation package called Merlion that increases the simulation speed of TEVASIM by about an order of magnitude for large numbers of contamination. This package is also used to efficiently generate an explicit mathematical model of the contaminant spread for use in mixed-integer programming formulations.

Second, we present a pair of mixed-integer programming (MILP) formulations for source inversion and optimal sampling. The source inversion problem involves locating the contamination source given a small set of measurements. Advancing existing techniques, we have developed an MILP formulation that makes use of discrete (yes/no) measurements from both fixed sensors and manual grab samples to successfully invert for the potential contamination sources. Using an integrated strategy, we then solve another MILP formulation to find the optimal manual grab sample locations for the next sampling cycle. This integrated strategy is effective in finding the contamination source using a small number of sampling cycles.

While identifying the contamination source, population exposure is continuing. One measure to mitigate this exposure is to increase chlorination of the system from a set of disinfectant booster stations. Finally, we propose a MILP formulation for optimally placing booster stations to help neutralize possible contamination events. Due to the large uncertainty in the contaminant injection location and time, this placement problem is a stochastic problem that would normally be solved using decomposition techniques. However, we propose a formulation and solution approach that is both fast and memory efficient. We use Merlion to rapidly solve the large number of possible contamination scenarios and formulate the stochastic programming problem. Using a set of reduction techniques that combine scenarios, but yield an exact mathematical transformation of the original stochastic problem, we can dramatically reduce the size of the complete stochastic programming problem. This approach has been used to place booster stations in a network with more than 3000 nodes and a stochastic programming formulation with more than 400,000 scenarios.

Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000.