Regan Murray

Review of Sensor Placement Strategies for Contamination Warning Systems in Drinking Water Distribution Systems

Contamination warning systems (CWSs) are a promising approach for the mitigation of contamination risks in drinking water distribution systems. A critical aspect of the design of a CWS is the strategic placement of online sensors that rapidly detect contaminants. This paper reviews the array of optimization-based sensor placement strategies that have been recently proposed. These strategies are critiqued and several key issues are identified that need to be addressed in future work.

Greedy Heuristic Methods for Locating Water Quality Sensors in Distribution Systems

Monitoring and surveillance systems for drinking water distribution networks are intended to provide real time warning of drinking water contamination events and mitigate their public health consequences. Drinking water distribution networks often serve large populations over vast areas. There exist a large number of access points where contaminants could be introduced, and these are spread throughout the service area. Transport of contaminants from these access points to consumers would occur through a multitude of pathways, and be dominated by water flows that change magnitude and direction in response to frequent changes in water use and system operation. The above features of drinking water distribution networks dictate that design of a successful monitoring and surveillance system is comprised of three interrelated sub-tasks:

FORMULATION AND OPTIMIZATION OF ROBUST SENSOR PLACEMENT PROBLEMS FOR CONTAMINANT WARNING SYSTEMS

The sensor placement problem (SPP) in contaminant warning system (CWS) design for water distribution networks involves maximizing the level of protection afforded by a limited number of sensors. In existing SPP formulations, the protection level is typically quantified as either the expected impact of a contamination event, weighted by occurrence probability, or the proportion of events that are detectable. In these formulations, the issue of how to mitigate against potentially high-impact events is either handled implicitly or ignored entirely. Consequently, any solutions of these formulations run the serious risk of failing to protect against any number of high-impact, 9/11-style attacks. This risk is further amplified by the fact that reliable estimation of contamination event probabilities is extremely difficult, such that existing SPP formulations may significantly discount the potential of high-impact events. In contrast, robust formulations of the SPP directly address these concerns by focusing strictly on a subset of high-impact contamination events, and placing sensors to minimize the impact of such events. We introduce several robust formulations of the SPP that are distinguished by how they quantify the potential damage due to high-impact contamination events. These include minimization of the worst-case impact, the Value at Risk (VaR), and the Tail-Conditional Expectation (TCE). The worst-case formulation is equivalent to the p-center problem in facility location theory. VaR and TCE are standard measures of robustness in the financial literature; the corresponding robust formulations of the SPP respectively minimize the (1-α)% largest impact and a weighted sum of the α% largest impacts. All formulations can be expressed as Mixed-Integer Programs (MIPs), which can be solved using both commercial MIP solvers and specialized heuristics. Additionally, we develop computational methods for exploring the performance trade-offs between robust and expectation-based SPP formulations. We use this framework to explore the nature of robust versus expectation-based solutions to the SPP on three real-world water distribution networks, ranging in size from 400 to over 10,000 junctions. We observe that robust SPP formulations are one or more orders of magnitude more difficult to solve than expectation-based SPPs. Our results indicate that simple heuristics yield optimal solutions to the smaller test problems in shorter run-times than MIP solvers, and yield higher-quality solutions for larger test problems. For realistic sensor budgets, solutions with low expected impact fail to protect against large numbers of high-impact contamination events (with impact 5-10 times larger than the expectation). In contrast, we show that solutions to robust SPPs yield 10-25number and magnitude of high-impact events. In general, our results indicate that it is possible to trade off mean impact versus high impact performance

US Environmental Protection Agency Uses Operations Research to Reduce Contamination Risks in Drinking Water

The US Environmental Protection Agency (EPA) is the lead federal agency for the security of drinking water in the United States. The agency is responsible for providing information and technical assistance to the more than 50,000 water utilities across the country. The distributed physical layout of drinking-water utilities makes them inherently vulnerable to contamination incidents caused by terrorists. To counter this threat, the EPA is using operations research to design, test, and deploy contamination warning systems (CWSs) that rapidly detect the presence of contaminants in drinking water. We developed a software tool to optimize the design process, published a decision-making process to assist utilities in applying the tool, pilot-tested the tool on nine large water utilities, and provided training and technical assistance to a larger group of utilities. We formed a collaborative team of industry, academia, and government to critique our approach and share CWS deployment experiences. Our work has demonstrated that a CWS is a cost-effective, timely, and capable method of detecting a broad range of contaminants. Widespread application of these new systems will significantly reduce the risks associated with catastrophic contamination incidents: the median estimated fatalities reduction for the nine utilities already studied is 48 percent; the corresponding economic-impact reduction is over

The TEVA-SPOT Toolkit for Drinking Water Contaminant Warning System Design

19 billion. Because of this operations research program, online monitoring programs, such as a CWS, are now the accepted technology for reducing contamination risks in drinking water.

A Model for Estimating the Acute Health Impacts resulting from Consumption of Contaminated Drinking Water

We present the TEVA-SPOT Toolkit, a sensor placement optimization tool developed within the USEPA TEVA program. The TEVA-SPOT Toolkit provides a sensor placement framework that facilitates research in sensor placement optimization and enables the practical application of sensor placement solvers to real-world CWS design applications. This paper provides an overview of its key features, and then illustrates how this tool can be flexibly applied to solve a variety of different types of sensor placement problems.

MODELING AND SIMULATION OF ARSENATE FATE AND TRANSPORT IN A DISTRIBUTION SYSTEM SIMULATOR

SIR models predict the spread of disease over time through susceptible, in- fected, and recovered populations, and are often used to design public health intervention strategies. A modified SIR model is linked to flow and transport models for distribution systems in order to predict the health risks associated with contamination events. The pro- posed framework provides information about the spatial and temporal distribution of health risks in distribution systems, and is useful for understanding the vulnerability of distribution systems to contamination, but also for designing strategies to reduce risks.

COMBINING WATER QUALITY AND OPERATIONAL DATA FOR IMPROVED EVENT DETECTION

Multi-species water quality models can be used to predict the fate and transport of contaminants such as arsenic in water distribution networks. In recent work, water quality models have been used to simulate hypothetical contamination events, estimate potential human health effects, and characterize the ability of sensors to detect contamination. Little work has been done to calibrate water quality models and validate them against experimental data generated in Distribution System Simulators (DSSs). In this paper, results are reported from bench scale and pilot scale experiments performed with a DSS at U. S. EPA’s Test and Evaluation Facility in Cincinnati, Ohio. The parameters for a reversible adsorption model were estimated from bench scale data generated over two days. The model was used with the EPANET-MSX software package to simulate the pilot scale experiment in the DSS. Model results match the pilot scale data very well for the first two days after the arsenate injection, however pilot scale data after this time deviates from model predictions. This deviation may be due to limitations in the time scale or sample size of the bench scale experiment. Additional modeling, simulation, and experimental work is planned to develop a fate and transport model that can be used in practical settings to design decontamination strategies following intentional arsenic contamination of water distribution systems.

FORMULATING AND ANALYZING MULTI-STAGE SENSOR PLACEMENT PROBLEMS

Water quality signals from sensors provide a snapshot of the water quality at the monitoring station at discrete sample times. These data are typically processed by event detection systems to determine the probability of a water quality event occurring at each sample time. Inherent noise in sensor data and rapid changes in water quality due to operational actions can cause false alarms in event detection systems. While the event determination can be made solely on the data from each signal at the current time step, combining data across signals and backwards in time can provide a richer set of data for event detection. Here we examine the ability of algebraic combinations and other transformations of the raw signals to further decrease false alarms. As an example, using operational events such as one or more pumps turning on or off to define a period of decreased detection sensitivity is one approach to limiting false alarms. This method is effective when lag times are known or when the sensors are co-located with the equipment causing the change. The CANARY software was used to test and demonstrate these combinatorial techniques for improving sensitivity and decreasing false alarms on both background data and data with simulated events.

Comparing Single- and Multi-Species Water Quality Modeling Approaches for Assessing Contamination Exposure in Drinking Water Distribution Systems

The optimization of sensor placements is a key aspect of the design of contaminant warning systems for automatically detecting contaminants in water distribution systems. Although researchers have generally assumed that all sensors are placed at the same time, in practice sensor networks will likely grow and evolve over time. For example, limitations for a water utility’s budget may dictate an staged, incremental deployment of sensors over many years. We describe optimization formulations of multi-stage sensor placement problems. The objective of these formulations includes an explicit trade-off between the value of the initially deployed and final sensor networks. This trade-off motivates the deployment of sensors in initial stages of the deployment schedule, even though these choices typically lead to a solution that is suboptimal when compared to placing all sensors at once. These multi-stage sensor placement problems can be represented as mixed-integer programs, and we illustrate the impact of this trade-off using standard commercial solvers. We also describe a multi-stage formulation that models budget uncertainty, expressed as a tree of potential budget scenarios through time. Budget uncertainty is used to assess and hedge against risks due to a potentially incomplete deployment of a planned sensor network. This formulation is a multi-stage stochastic mixed-integer program, which are notoriously difficult to solve. We apply standard commercial solvers to small-scale test problems, enabling us to effectively analyze multi-stage sensor placement problems subject to budget uncertainties, and assess the impact of accounting for such uncertainty relative to a deterministic multi-stage model.