Agent simulation for prediction of post-earthquake mass evacuation

An agent simulation is developed to predict unorganized emergency evacuation process of crowds consisting of various people. An agent is able autonomously to see, think and move; the distribution of its speed is the same as the walking speed distribution which is measured by analyzing video images of people running in an emergency state, and a model for an evacuation space is automatically constructed from digital data which are stored in GIS or CAD. The simulation is carried out to a network of narrow streets, subway stations and a large underground shopping mole. It is shown that the developed agent simulation is able to simulate unorganized state of mass evacuation and to analyze risk of the evacuation space.