Parallel real time computation of large scale pedestrian evacuations

Usually, modeling of the evacuations is done during the planning and authorizing process of office buildings or large scale facilities, where computing time is not an issue at all. The collaborative Hermes project [1] aims at improving the safety of mass events by constructing an evacuation assistant, a decision support system for heads of operation in an actual evacuation. For this, the status (occupancy and available egress routes) of a facility is constantly monitored with automatic person counters, door sensors, smoke sensors, and manual input from security staff. Starting from this status, egress is simulated faster than real time, and the result visualized in a suitable fashion to show what is likely to happen in the next 15min. The test case for this evacuation assistant is the clearing of the ESPRIT Arena in Dusseldorf which holds 50,000-65,000 persons depending on the event type. The on site prediction requires the ability to simulate the egress in ~2min, a task that requires the combination of a fast algorithm and a parallel computer. The paper will describe the details of the evacuation problem, the architecture of the evacuation assistant, the pedestrian motion model employed and the optimization and parallelization of the code.