Crowd management for large-scale outdoor events: Multi-agent based modeling and simulation of a crowd behaviors

Crowd modeling and simulation have become an attractive scientific research area, especially when public safety has never been so vulnerable than before. Due to the complexity of crowd behavior, the pedestrian dynamics in large-scale outdoor events have not caught much attention and we have not had enough knowledge to answer the question "How to efficiently manage large number of people moving in a short period?". Applicable tools that can facilitate crowd management in such events are also absent in the integration of pedestrian's spatio-temporal behavior and their interactions with their behavioral environment. Considering the characteristics of collaborative works in management, the main objective of this dissertation is to set up a framework of geo-collaborative crowd management for large-scale outdoor events. Its focus is on modeling and simulating crowd behavior by the multi-agent approach.; Firstly, a conceptual framework for a geo-collaborative management of crowds in large-scale outdoor events has been proposed. This framework is developed on the data environment which stores the information and knowledge of crowd behavioral dynamics as well as their behavioral space. It also provides a visualization environment with multi-modal and multi-mode interfaces for different end-users to visualize and simulate processes of crowd behavior and events evolvement, and to facilitate communications and decisions-making. This proposed framework has shed light on the vision of moving crowd management towards geo-collaboration.; Secondly, through the study on the case of "TST Firework Display", crowd behaviors in outdoor events have been explored and analyzed from an aggregate perspective. Driving forces for crowd dynamics are comprehended by emphasizing the effects of urban spatial morphology on crowd behaviors. This urban effect is represented through pedestrian visibility differences produced by urban terrain and road networks. In this study, the on-site observation results and the analysis on historical events were used for summarizing the behavioral dynamics of outdoor crowds, enriching the knowledge on crowd behavior, and providing the basis for modeling crowd behavior in large-scale outdoor events.; Thirdly, multi-agent based pedestrian models for large-scale outdoor events (MAPMODE) are developed from a disaggregate perspective. In MAPMODE, each pedestrian is viewed as an intelligent agent. Due to their different roles, attractors, policemen, buildings, roads, facilities, and emergencies are also represented by different types of agent in the models. Three specific models are developed based on the phases formed, including the model for crowd arrival, the model for crowd dispersal and the model for crowd evacuation. Not only the long-term observations and experiences on crowd behavior, but also processes of pedestrian decision-making and spatial cognition of human behavior are combined in MAPMODE.; Finally, the platform of integration and simulation of crowd behavior (ISCB) is developed. Ten scenarios of the event are generated to represent crowd behavior under different conditions. It is found that multi-agent based modeling brings additional information on crowd behavior that can not easily be observed or estimated only through collective observations and analysis. Therefore, ISCB with the basic requirements of geo-collaborative crowd management framework provides the foundation to accomplish real geo-collaboration at same/different time and same/different places.