Multiscale Simulation of Pedestrians for Efficient Predictive Modeling in Large Events
Mohcine Chraibi and Bernard Steffen
The Hermes project  demonstrated the usefulness of on site predictive simulations of probable evacuation scenarios for security personnel. However, the hardware needed was prohibitively expensive . The present paper shows that a multiscale approach can perform the simulation in a fraction of time without loss of useful information. The main problem is the correct passing of agents from a coarse scale model to a fine scale model, here from a CA model to a force based model. This will be achieved by inserting agents into the force based model at positions and speeds optimized for smooth walking either by a priori information or using Voronoi cells. The resulting model is easily parallelised using OpenMP, achieving reasonable efficiency.
We also show how a CA method can be modified to address the problem better than standard CA, too. However, this approach needs further development and calibration to be reliable.
Keywords: Cellular automata, modeling, pedestrian dynamics