A Combined Netflow-Driven and Agent-Based Social Modeling Approach for Building Evacuation

In an emergency, finding safe egress pathways in a short period of time is crucial. In this paper we use a network flow (netflow) algorithm that acts as the core of a real-time recommender system to be used by building occupants and decision-making bodies. However, a purely optimization approach can lack realism since building occupants may not evacuate immediately, stopping to look for their friends or trying to assess if the alert is for real or just a drill, etc. Furthermore, they may not always follow the recommended optimal paths. Thus, in order to assess the egress in a physical space and to test our evacuation algorithms, we use a simulation-optimization (S/O) approach. The model allows us to test more realistic evacuation scenarios and compare them with an optimal approach. The S/O uses both a netflow algorithm and an agent-based approach to model and simulate individual human behaviours. People are modeled as agents with specific characteristics, such as social attachment to others, variation in speed of movement, etc. Furthermore, a Belief-Desire-Intention (BDI) agent architecture is used to model the individual differences in people and to more accurately describe the heterogeneity of the building occupants in terms of their current beliefs about the situation and goals. The real geospatial data obtained from three experiments is set as the model input. The results confirm the usefulness of using such S/O approach to improve design-time and real-time evacuation systems.