マルチエージェントシステムによる歩行者行列の再現と追従行動が及ぼす群集流動の分析

Abstract

A pedestrian simulation is one of effective means to verify the size of space and share the information of a crowd safety measure in an urban area that a large number of unspecified people visit. This study aims to reproduce pedestrian movements in high density queues and to clarify a mechanism of lane formation in contra-flow by using a multi-agent system. The crowd flow is based on the social force model, which is said that crowd dynamics at a state of emergency can be reproduced. To apply this model to pedestrian movements in a high density queue at ordinary times, we add new functions, which are ellipse-based collision detection and following behavior after an agent ahead. The following behavior includes the parameters to control propagation speed of stop-and-go waves in a queue. To confirm the reproducibility, we compare observation data of pedestrian movements in queues at urban facilities. In addition, we simulate a contra-flow and the lane formation phenomena to analyze the influence of the following behavior on crowd dynamics. The results are as follows. If the shape of an agent is a circle of the radius 30 cm, the collision between agents is detected at the line density of 1.67 people per meter. On the other hand, the system represents a pedestrian movement having the line density of 2.1 people per meter by making the shape of an agent an ellipse 30 cm in the major axis and 15 cm in the minor axis. In addition, by dynamic control of the parameter values (a, b) involving the fundamental diagram for pedestrians, and the parameter values (Ci, Di) involving line density during standing and walking, it is possible to reproduce the characteristic values of observation data, which are line density, propagation speed of stop-and-go waves. If the static parameter values are set, it is hard to agree with propagation speed of several stop-and-go waves accurately. However, it is useful to predict the characteristics of pedestrian movements in a queue. As a result of an application to contra-flows by using the multi-agent system, the lanes are formed regardless of the following behavior when the crowd density is low. On the other hand, the following behavior facilitates lane formation in the contra-flow with a density of over 1.0 people per square meter. It is an important point to enhance crowd safety. From the above results, we developed the general-purpose multi-agent system which enables to reproduce pedestrian movements in a queue and lane formation of contra-flow by extending an existing crowd dynamics model at a state of emergency.