Agent-based fire evacuation model using social learning theory and intelligent optimization algorithms

Abstract

Fire incidents often lead to a series of social problems. Therefore, it is particularly important to optimize evacuation strategies and promote relevant social safety knowledge. Based on this, the study proposes a fire evacuation model that integrates the Fire Dynamics Simulator (FDS) with Agent-Based Modeling (ABM) to simulate a bar fire scenario. In this model, the concept of social learning is introduced, and multiple factors such as evacuation time, trampling risk, and pedestrian health are considered as risk evaluation indicators. Machine learning combined with intelligent optimization methods is applied to optimize evacuation strategies. First, we validate the effectiveness of the model by comparing the averaged simulation results with real-world data. The results demonstrate that the simulation outcomes of our model exhibit good accuracy and robustness. Secondly, we analyze the importance of the second-floor safety exit. When the second-floor safety exit remains unobstructed, evacuation efficiency and casualty risk can be significantly improved. Then, we examine the role of social knowledge. When people are aware of the fire risk and choose to evacuate immediately, casualties can be significantly reduced. Finally, we study the effectiveness of phased evacuation in enhancing crowd safety. By employing a method that combines Random Forest and the Particle Swarm Optimization-Genetic Algorithm (PSO-GA), phased evacuation strategies are optimized, resulting in definitive strategies to reduce evacuation risks. This finding further expands social knowledge, indicating that when the proportion of staggered evacuation is appropriate, evacuation risks can be significantly reduced. Our research contributes to the development of social safety knowledge and provides methodological references for formulating evacuation strategies in different settings.