Thesis (M.S., Computer Science) -- University of Idaho, 2017 | Evacuation planning is a fundamental part of emergency management. An effective evacuation plan can mitigate significant loss of life. During a disaster many factors can complicate evacuation, such as traffic congestion, real-time damage to the transportation network, changes in the safety of areas in the city, and noncompliance with established evacuation plans. Previously, traffic in an evacuation has been managed as a routing problem from origins to destinations.
This work describes an evolution-based approach to the problem of evacuation planning using a Markov model approach. It is designed to better adapt to unpredictable complications of disaster evacuation. In our approach an Evolution Strategies algorithm is applied to sets of probabilities describing assignment of traffic to streets in networks representing urban areas. A mesoscopic traffic simulation is used to evaluate the fitness of each set of probabilities, establishing a traffic assignment distribution. Fitness is evaluated by measuring the safety of all vehicles at the end of the simulation. This method allows abstraction of individual vehicles and origin-destination pairs, allowing a more generalized solution. Sources of danger creating the need for evacuation are also abstracted, allowing the application of this model to arbitrary disaster events.