Developing an Adaptive Evacuation Simulation Framework based on Agent Performance Dynamics

Abstract

The benefits of evacuation drills for disaster preparedness are not yet well-established. Virtual reality and serious games (SG) application to evacuation simulation were proven to be effective in overcoming limitations of traditional emergency training approaches with a significant increase in the trainees' knowledge and self-efficacy. Further, human behavioural modelling is a critical component of evacuation simulation. This study will develop an evacuation simulation framework, incorporating various approaches such as cluster analysis, structural equation modelling, agent-based modelling, cognitive modelling and serious games. Although the case study focuses on earthquake, the components of the evacuation framework is designed to be applicable to other types of disaster requiring evacuation. The study's contributions include 1) a cognitive agent model, incorporating experiences, personality, emotions and behavioural decision-making, 2) an implementation of the behavioural modification approach as the foundation for efficient evacuation, 3) a reproduction of a multi-storey structure to add layers of complexity to the evacuation, and 4) a decision support tool to evaluate existing strategies and propose new strategies that will adapt to agent performance dynamics. The study site is a university with a student population of approximately 8000, located at around 8.4 km from a major earthquake fault line and lacks open spaces for evacuees and emergency responders. Therefore, the site is significantly at high risk for a large number of casualties once the anticipated earthquake of extreme magnitude with a minimum of 7.2 strikes. Four scenarios are being considered including Scenario 1 which represents a regular enrolment day when student numbers are expected to be moderately higher than a regular school day. Scenario 2 represents the peak time slot on a regular school day when the population is at its maximum. Scenarios 3 and 4 are both days with extra-curricular events when the number of visitors is at its maximum. An online earthquake evacuation survey was designed to conduct population profiling for the occupants of the university. Cluster analysis will divide the survey participant population into evacuee groups in terms of their demographic information, personality traits and their responses to questions on past and future behaviour and decision-making during evacuations. Preliminary implementation in Unity game development software involved simulation runs using an initial version of the agents being aware of the exit locations. For each timestamp, the simulation engine stores information per agent including a generated unique agent number ID, role including faculty and student, current location such as first floor and parking lot, 3D coordinates and the evacuation state. The evacuation simulation component will produce evacuee count vs time and evacuee floor level vs time graphs, and evacuee route and heat maps. The decision support component will produce a set of proposed evacuee schedule and route assignments. The proposal for the evacuee population profile composition will be based on the experiments that will be conducted focusing on how behavioural modification can improve evacuation performance. The succeeding version of the implementation will apply the results of the survey on the agent model and perform the simulations on the four identified crucial scenarios.