Simulation Software Selection and Model Validation Standards in Crowd Evacuation

Problem Description

In crowd evacuation research, simulation software is a core tool for modeling evacuation processes and evaluating the effectiveness of strategies. Different software (such as FDS+Evac, Pathfinder, AnyLogic, etc.) vary in modeling principles, applicable scenarios, and accuracy. This task requires analyzing how to select simulation software based on specific needs and establish scientific model validation standards to ensure simulation results are close to reality.

Problem-Solving Process

1. Define Simulation Objectives and Requirements

Software selection should be based on the needs of the specific scenario. Key questions include:

• Evacuation Scale: Is it a small-scale (e.g., single-story building) or a large-scale (e.g., subway station, stadium) evacuation?
• Behavioral Complexity: Is it necessary to simulate individual decision-making differences, social relationships, or panic spread?
• Output Metrics: Is the focus on evacuation time, congestion point analysis, or path efficiency optimization?
• Data Support: Are there field observation data (e.g., video trajectories, flow statistics) available for model calibration?

Examples:

• For high-fidelity physical simulation (e.g., the impact of fire spread on evacuation), FDS+Evac (based on fluid dynamics and social force models) is an option.
• For emphasizing behavioral diversity (e.g., family evacuation, guide-led evacuation), AnyLogic (supporting multi-agent systems and rule customization) is preferable.
• For rapidly simulating macroscopic flow of large crowds, Pathfinder (based on grids and navigation meshes) is more efficient.

2. Compare Core Modeling Principles of Software

The underlying logic of different software directly affects its applicability:

Key Points:

• Social Force Models (e.g., FDS+Evac) are more suitable for simulating pushing, shoving, and congestion in high-density scenarios but require high computational costs.
• Rule-Based Models (e.g., Pathfinder) rely on predefined path selection logic, offering high efficiency but may overlook emergent behaviors.
• Multi-Agent Models (e.g., AnyLogic) can integrate psychological rules (e.g., herd behavior) but require extensive parameter calibration.

3. Establish Model Validation Standards

Simulation results must be validated before being used in practical decision-making. Common standards include:

(1) Internal Validation (Verification)

• Purpose: Ensure the model code operates as intended and eliminate program errors.
• Methods:
  • Extreme Scenario Testing: For example, testing an empty scenario to verify if evacuation time approaches zero.
  • Parameter Sensitivity Analysis: Adjust key parameters (e.g., walking speed) and observe if output changes are reasonable.
  • Grid Convergence Test: In grid-based models like Pathfinder, reduce grid size to check if results stabilize.

(2) External Validation (Validation)

• Purpose: Compare simulation data with real-world data to ensure the model reflects reality.
• Methods:
  • Macro-Indicator Comparison: For example, ensuring errors in total evacuation time or bottleneck flow rates do not exceed 15% compared to field observations.
  • Micro-Trajectory Analysis: Calibrate individual path choices and speed distributions using video data (e.g., Kolgomorov-Smirnov test).
  • Special Behavior Reproduction: Check if behaviors like逆行 (counter-flow) during panic or family clustering are reasonably simulated.

(3) Calibration

• If initial validation fails, iteratively adjust parameters:
  • Key Parameters: Walking speed distribution (normal distribution mean 1.2-1.5 m/s), decision delay time (0.5-2 s).
  • Optimization Tools: Use genetic algorithms or Bayesian calibration to minimize differences between simulation and real data.

4. Case Study: Stadium Evacuation Simulation Selection

• Requirements: Large-scale (50,000 people), multiple exits, evaluation of signage effectiveness.
• Software Selection:
  • Exclude FDS+Evac (excessive computational resource requirements).
  • Pathfinder is suitable for rapid testing of exit layouts but requires supplementary behavioral rules (e.g., spectators navigating around seats).
  • Ultimately, AnyLogic was chosen, with custom parameters like "area familiarity" to simulate path differences between local spectators and tourists.
• Validation Process:
  • Internal Validation: Verify that navigation meshes cover all stand aisles.
  • External Validation: Compare with historical drill data to calibrate clustering behaviors of family groups.
  • Results: Simulation and measured evacuation time error ≤12%, with congestion point prediction accuracy exceeding 80%.

Summary

Selecting simulation software requires balancing modeling objectives, data resources, and computational constraints. Model validation must be ensured through internal logic checks, external data comparisons, and parameter calibration. In practice, a complementary multi-software strategy (e.g., using Pathfinder for preliminary layout optimization and AnyLogic for detailed behavioral analysis) is often adopted to enhance the reliability and practicality of results.