Crowd flow forecasting via agent-based simulations with sequential latent parameter estimation from aggregate...

Area of Science:

• Computational Social Science
• Urban Planning and Management
• Data Science and Simulation

Background:

• Real-time crowd simulations are crucial for smart city management but forecasting large crowds remains challenging.
• Existing methods struggle with microscopic interactions and large-scale crowd flow prediction.
• Incorporating real-time observations into simulations is an active research area.

Purpose of the Study:

• To develop a method for real-time crowd flow forecasting using microscopic agent-based simulations.
• To accurately predict crowd movements for thousands of individuals by integrating observation data.
• To enhance crowd management strategies in smart cities through improved forecasting.

Main Methods:

• Utilized a microscopic agent-based model for crowd simulation.
• Employed a particle filter algorithm to sequentially estimate crowd state and latent parameters.
• Integrated aggregate crowd density observations into the simulation framework.
• Validated the method with numerical experiments, including a large-scale evacuation scenario.

Main Results:

• Successfully forecasted crowd flow for large crowds (thousands of individuals).
• Demonstrated reasonable accuracy in predicting crowd movements even with limited observation data.
• Showcased the method's effectiveness in a realistic evacuation scenario.
• Validated the feasibility of real-time forecasting for microscopic crowd dynamics.

Conclusions:

• The proposed method enables accurate real-time crowd flow forecasting for large populations.
• This approach supports enhanced crowd management in smart cities by providing predictive insights.
• Agent-based modeling combined with observation data offers a viable solution for complex crowd dynamics.