Modeling and simulation of multi-agent systems for emergency scenarios

The evacuation behavioral model of human beings requires a more comprehensive approach in emergency situations. This thesis describes the proposed human behavior system that integrates both artificial intelligence and fuzzy logic parameters. Fuzzy logic is used in refining human emotional behavior by including fuzzy features such as stress, panic, smoke and anger. Humans are unpredictable when it comes to making decisions during emergency situations. This unpredictability is due to human behavior characteristics. The incorporation of the human behavioral model through the use of fuzzy parameters brings a sense of reality. This dissertation describes a new approach to model human behavior characteristics and represents them in a goal finding multi-agent environment. This thesis proposes an algorithm for evacuation scenarios based on geometric approach that realistically models the behavior of people using the choice of exit (favorite goal) and sub goals (intermediate goals). The simulation has agents following a safe trajectory and making appropriate decisions to avoid 198 collisions with static elements of the physical environment as well as with other mobile agents. The proposed research tool AvatarSim is developed in Java framework and can be displayed on a web browser. The model is extended to simulate battle field scenarios and airplane evacuation scenarios. The tool can be programmed with models that can be informed by an analysis of evacuation procedures and also by observations of human behavior during a real-time evacuations scenario. This thesis also proposes a fuzzy and neuro-fuzzy application for predicting the probability of an agent to reach its destination when fuzzy behavior characteristics like stress and panic are involved. The fuzzy system allows easy encoding of expert knowledge as a set of rules. The validation of the model was done through evacuation experimental studies performed at Wayne State University for comparison of proposed AvatarSim model predictions with reliable information from a real time evacuation drill. The results from the study prove that the evacuation time and flow rate of people were comparatively similar for AvatarSim when compared to real-time evacuation drill.