(32cp) Agent-Based Modelling for Evacuation in Industrial Settings: A Stochastic Approach for Critical Scenarios

Industrial facilities are often complex environments with high risks due to the presence of hazardous materials, equipment, and confined spaces. Efficient and safe evacuation planning is of significant importance, especially under emergency scenarios where rapid response can prevent casualties and mitigate damages. Traditional evacuation models may lack the flexibility to account for individual variability and dynamic conditions inherent in industrial settings. This study introduces a robust agent-based modeling (ABM) framework designed to simulate evacuation processes in industrial environments, accounting for random elements and critical scenario-specific factors as well as multiple scenarios applications based on 3D representations of industrial installations.

The primary objective of this study is to develop and validate a stochastic ABM that accurately simulates evacuation behavior in industrial settings, particularly in scenarios with high uncertainty, such as fires, chemical spills, or explosions. The model seeks to improve understanding of evacuation dynamics by considering individual agent characteristics, environmental obstacles, human being factors, high stress situations, and varying levels of hazard exposure. The model incorporates a virtual industrial layout, which can be adjusted given 3D representations, defined by physical barriers, hazard zones, and emergency exits. Agents represent individual workers with attributes that affect movement, such as age, familiarity with the layout, and mobility limitations. Each agent’s behavior is influenced by stochastic elements, including random movement delays, decision-making variability, high stress levels, and interaction with hazards. Evacuation paths are optimized through shortest-path algorithms, while interactions between agents and obstacles are handled using a collision-avoidance system. Model parameters were calibrated using empirical data from historical evacuation drills. Multiple scenarios were simulated, including partial and full evacuations, to examine model robustness and adaptability under different hazard intensities, scenarios, and agent behaviors.

Simulations revealed that stochastic agent behaviors significantly impact evacuation times, with scenarios involving high levels of hazard exposure resulting in non-linear increases in evacuation time. The model showed that in critical scenarios, such as unexpected hazard spread, agents with high familiarity and mobility had shorter evacuation times. Comparative analysis demonstrated that the ABM approach provided more realistic evacuation timelines than traditional deterministic models, particularly when simulating panic and congestion in high-density areas. Sensitivity analysis identified key factors such as exit accessibility and individual agent risk tolerance that strongly influence evacuation efficacy.

This stochastic ABM framework offers a powerful tool for planning and optimizing evacuation strategies in industrial settings, addressing the limitations of deterministic models by incorporating individual-level variability and environmental dynamics. ABM adaptability to numerous emergency scenarios provides valuable insights for facility design, emergency preparedness, and worker training. Future work would extend this model by integrating real-time hazard detection data using Internet of Things sensor infrastructure and further refining agent decision-making processes, ultimately contributing to safer industrial environments and improved risk management strategies and mitigation.