A novel data-driven multi-agent pedestrian flow risk assessment framework for avoiding stampede incident

Area of Science:

• * Computational Social Science
• * Urban Planning and Safety
• * Artificial Intelligence in Crowd Management

Background:

• * Existing crowd simulation tools struggle with real-world complexities like varied destinations and multi-directional flows.
• * Monitoring high-density crowds in public spaces is crucial for preventing accidents and ensuring safety.
• * Communal areas often feature complex spatial designs that challenge current monitoring methods.

Purpose of the Study:

• * To introduce a novel data-driven, multi-agent framework for assessing crowd dynamics and early warning conditions.
• * To develop a system capable of handling diverse spatial layouts and pedestrian behaviors.
• * To enhance the prediction and prevention of overcrowding incidents in public spaces.

Main Methods:

• * Utilizes real-time visual information and reinforcement learning for intelligent decision-making.
• * Employs a self-iterative algorithm for trajectory planning that mimics real-world pedestrian movement.
• * Ensures model adaptability across various scenarios without requiring parameter fine-tuning.

Main Results:

• * The framework accurately reproduces pedestrian flow in diverse, complex scenarios.
• * Identified a discontinuous state transition in pedestrian flow as density increases.
• * Proposed a method to detect building traffic capacity and establish crowding warning thresholds.

Conclusions:

• * The developed framework effectively assesses crowd dynamics and predicts early warning conditions.
• * Rational spatial layout and information guidance can significantly improve mobility and reduce stampede risks.
• * The approach offers a method for proactive crowd safety management without architectural expansion.