Combining Simulation and Mean Field Analysis in Quantitative Evaluation of Crowd Evacuation Scenarios

Crowd movement analysis methods structurally suffer from the problems of scalability and lack of empirical data. Macro scale approaches can tackle larger crowds, but they prevent direct representation of dynamics occurring on the micro scale which may be more easily related to observations. In this work, we present and experiment a hierarchical approach which aims at conciliating the contrast through the combination of fine grained simulation and coarse grained analytical techniques. To this end, agent based simulation is applied on micro scale patches where mechanisms resembling reality can be more easily reproduced. Measurements on simulation results are then used as parameters for an analytical model which exploits mean field techniques to efficiently approximate the emerging crowd behavior through the fluid limit obtained when the crowd density tends to infinity. The approach is experimented with reference to a crowd evacuation scenario, using NetLogo as engine for spatial agent based simulation and JSam framework as a solution engine for mean field analysis.