Agent-Based Modelling and Simulation of Pedestrian Evacuation Behaviour during Large Outdoor Events---Fireworks Display at Tsim Sha Tsui of Hong Kong as a Case Study.

摘要

Public safety has become more and more important nowadays as tragic incidents, such as the 9/11 World Trading Centre attack, have caused a large number of causalities. It is critical to improve the capability of the security department for safely managing the crowd in response to an emergency. Conventional techniques investigating evacuation dynamics, such as the sand-box game utilized by Hong Kong police department preparing evacuation plans for the case event, are generally inadequate for providing sufficient support to crowd management. Enhancing the management practice by computer based simulation techniques has been attracting an increasing amount of research interests recently, as it can offer timely decision making assistance in changing situations compared to the conventional approaches.;In reviewing the existing evacuation models and simulation systems, it was found that although there have been remarkable advances in reproducing and analyzing many phenomena and the dynamics of pedestrian flow in evacuations, social psychological factors were not well considered. Theories and research in the social psychological field have indicated that these factors are essential in evacuation behaviour. Therefore the author examined the socio-economic factors relating to psychological responses through a literature review and explored the potential for integrating such factors into the agent based modelling approach.;Inspired by such concerns, the author proposed the individual cognitive behaviour model consisting of the Personalized Spatial Cognitive Road Network, which represents the diverse individual spatial knowledge of recognizable roads, and the Multiple Routing Strategies which incorporated individual preferences in selecting routing strategy. The proposed model follows the principle that pedestrians are considered to be composed of heterogeneous individuals, thus each pedestrian could have his or her own knowledge base and rules of behaviour. Two steps in constructing each component were presented respectively, with an integrated framework presented in the context of agent based modelling.;In order to implement the model, the layered behavioural simulation model was proposed for putting the decisions made from the behaviour model into action. By identifying the three layers sustaining the routing procedure at their respective levels, the evacuation route was finally converted to selecting stepping cells from adjacent neighbours. In this way a quick way-finding process can be achieved toward the ultimate destination, so that the global objectives are coordinated with local action decisions on the cell scale. With this simulation model, the framework of the geo-referenced agent-based system was proposed with technical details presented. This system was built directly on a GIS base and thus was able to conduct simulation in a real space environment and perform advanced GIS analysis inside the simulation.;By conducting a field survey to collect and investigate the attributes and behaviour patterns of participants in the case event, the proposed model and simulation system were implemented. Three scenarios of different populations of involved agents were simulated on the finished system according to the sample data. Simulation results were analyzed and compared to the observed data. The simulation system was found to be capable of reproducing compatible dynamics of pedestrian flows.;Finally the conclusions of this thesis study were presented.