Incorporating groups, collective behavior, and information visualization in agent-based models of evacuation

Abstract

This dissertation is intended to advance research in building evacuation modeling through the introduction of detailed social groups, collective behavior, and improvements in information visualization. The model built as a part of this dissertation makes significant original contributions to both input and output of building evacuation models. ☐ Regarding inputs, this work prototypes new ways to catalog social groups, leadership, and the concept of supra force - a combination of high density, contraflows of crowds, and environment. The central difference between this model and previous efforts is the role of group affiliations. The effort resulting from this dissertation, SocEvac, creates a three-layer decision tree for most agents, who have to balance individual and group responsibilities while attempting to avoid supra force. This interaction of individually optimal exit paths and social responsibilities creates significantly more contraflow situations as agents attempt to locate and evacuate with their loved ones. These contraflows impede efficient evacuation, helping to explain scenarios such as the Station nightclub evacuation; where there were significantly more fatalities then would have been expected based on population and number of exits. ☐ On the output side, this work creates new methods to examine simulation models in real-time, and suggests new methods of measurement to determine what makes an accurate model. The real-time visualization methods allow for researchers to quickly understand what is happening while a model is running. These visualizations allow for users of a simulation to control what features they want to highlight in a model in real-time. The new methods of output measurement center around tracking agents as individuals, cataloging outcomes of agents each modeling a real-world counterpart complete with demographics and relationships. By transitioning away from aggregate population tracking and focusing on individuals, it is now possible to compare models to an entire evacuation narrative instead of only attempting to recreate end results. ☐ These improved inputs and outputs result in the most descriptive model to date of population movements during the Station nightclub fire in 2003. More than ten years after the fire, I believe the SocEvac model can finally begin to explain the complex events that led to the high fatalities and unconventional exit paths of the evacuation. ☐ While this dissertation focuses on one scenario, the underlying program can be used to model almost any building evacuation. This platform is designed to inspire other model builders to consider adding group social behaviors to models.