DECISION SUPPORT FOR CRITICAL INFRASTRUCTURE RECOVERY
Analytics
70 views ◎40 downloads ⇓
Abstract
Protecting critical infrastructure systems, such as electrical power grids, has become a primary concern for many governments and organizations across a variety of stakeholder perspectives. Critical infrastructures involve multi dimensional, highly complex collections of technologies, processes, and people, and as such, are vulnerable to potentially catastrophic failures and cascading effects with escalating impact across multiple infrastructures. Understanding the impact of service outages in utility services, including electric power, water, and natural gas, is a key part of decision making in response and recovery efforts. In this dissertation, I present a framework and prototype interactive geovisualization interface for a static critical infrastructure reconstitution task and showed it provides a natural context for this infrastructure analysis support. Results of my first experiment using the prototype for indicate that the decision makers can make better decisions with less time and less cognitive load using a software tool based on the decision recommender framework. Often, data that encapsulate the source-sink relationships between utility service points and customers are confidential or proprietary, and, therefore, unavailable to external sources due to their sensitive nature. As a result, during emergencies, external decision-makers often rely on estimations of service areas produced by various methods. During our decision support tools user study, critical infrastructure and geographic information analysis experts expressed a need for highly accurate estimates. Therefore, I tested traditional distance-based and cell-based methods for service area approximation methods on an electric power network for a mid–size Midwestern US city. Results from my second experiment showed that methods that take capacity and demand into account outperform standard methods. In addition, I also found that cell-based methods were more accurate when demand is closer to sources, indicating that cell-based methods may work best for large areas, such as states, while distance-based methods may work best for locations with uniform demand. When the desired analysis includes aggregate economic or population predictions, the weighted version of cellular automata (CA) performs best. When the desired analysis includes facility–specific predictions, then weighted Thiessen–based approaches tend to perform best. Based on insights from the electric power network service area research, I developed two novel service area estimation methods based on based on road network optimization. To further understand the relative merits of each method, I devised a novel adaptation of accuracy assessment methods from land cover classification to service area estimation. Results of my third experiment estimating water service areas for the state of Kentucky indicate that service area estimation methods based on road network optimization will produce more accurate results compared to distance or cell-based estimation methods. However, in my fourth experiment, these new methods did not compare well with traditional methods on the power network for a city, demonstrating that particular care is needed to ensure that approximation methods are chosen to align with the properties of the service network, the population distribution, and the available source and demand data.Results of the third experiment estimating water service areas for the state of Kentucky indicate that service area estimation methods based on road network optimization produce more accurate results compared to distance or cell-based estimation methods. However, the fourth experiment applying these methods to the electric power data for a city showed that traditional service area estimation methods outperform transportation-based methods. This is likely due to the absence of zoning or industrial demand area definitions in the reference dataset. This striking difference in results highlights the importance of the distribution of sources and sinks in critical infrastructures.This dissertation presents an overview of my approach and the GIS modeling environment for decision support in critical infrastructure reconstitution. In addition it provides insight into various service area approximation methods for increased recommendation accuracy. Results from the experiments performed and explained in this dissertation will be valuable when designing real world, smart, and accurate decision support systems.