Development of multi-agent simulation model for evacuation from landslide disaster area

Japan is a country which has many steep slopes along rivers. Therefore, the disasters due to landslide and mud flood have often occurred in mountainous urbanized areas. In rainy or typhoon season much rain triggers such disasters because of unusual rainfall. In these years those disasters have been increased in not only mountainous areas but also the housing area developed in the fringe of urban region newly. On the other hand, some of the nursing homes for aging people located in such disaster estimated areas. The people have also fallen victim to landslide disasters. So it has been important for a local government to build a comprehensive evacuation program of a large natural disaster. In this program, it is substantial to give appropriate information on human behavior for the evacuating time. In this study a new methodology based on behavior-oriented agent system should be discussed to develop. The production roles of the attributive groups were built in terms of the questionnaire survey on evacuation trips from damaged districts. Using the set of production rules composed of the questionnaire data, a multi-agent simulation model in a hypothetical landslide disaster was developed by a multi-agent system (MAS) method. An agent in this study is the person that can perceive its environment through sensors and decide the activities due to effectors. The multi-agent system was simulated in terms of evacuation from a damaged district to a safe shelter in Sapporo, Japan. It comes to the conclusion that the human behaviors and their interactions during the disaster impact were constructed by multi-agent simulation model and the possibility of the evacuation was found in view of both topographic and human attributive conditions.