Introducing uncertainty into evacuation modeling via dynamic traffic assignment with probabilistic demand and capacity constraints

Emergency evacuations are low-probability-high-consequence events that have attracted the attention of researchers since 1960s. An evacuation process can be triggered by various natural (hurricane, flood, tsunami etc.) and man-made (industrial accidents, terrorist attack etc.) events. Regardless of the threat, the nature of the evacuation process involves a very high utilization of the transportation network and searching for plans/strategies to move large number of people to a safe place in the shortest possible time. Researchers from different disciplines approach to the evacuation problem from different perspectives. Two major components of any evacuation event are estimation of the evacuation demand and traffic analysis to make planning inferences about the evacuation performance measures such as clearance time. Although related studies and real-life practices show a significant uncertainty regarding the evacuation demand due to the unpredictability of human behavior and changing roadway as a result of disaster impacts, the state-of-the-practice does not consider this type of randomness. This dissertation aims to address this important gap by proposing a dynamic traffic assignment formulation with probabilistic constraints that takes into account uncertainties in demand and roadway capacities. The proposed model uses a cell transmission model based system optimal dynamic traffic assignment formulation. The demand and roadway capacities are assumed to follow a discrete random distribution and the p-level efficient points approach [115] is employed to solve the proposed model. Two numerical examples regarding the use of the model are provided. The numerical examples also discuss the implications using individual chance constraints vs. joint chance constraints which provide different interpretations for the reliability of the results. Overall, the proposed formulation generates evacuation time performance measures that can be interpreted within reliability measures rather than single deterministic point estimates that would not be necessarily observed during a real life test, mainly due to high level of uncertainty created by human behavior and capacity impacts of the disaster.