Agent-based modeling of crowd dynamics in rail transit systems

Abstract

The rail transit systems in many modern cities, especially those in some Asian countries with dense populations, have been developing at an unbelievable speed recently. But with more and more people swarming into the rail transit systems, many metro stations and trains have become too crowded and it has become a big challenge to guarantee a safe and comfortable traveling environment for passengers. On the one hand, station designers and operators need new technologies to make appropriate decisions and strategies to keep the crowding level acceptable or sometimes revamping work for station layout will be needed. On the other hand, the performance of the station facilities and evacuation plans under emergency situations such as fire or rail accidents in tunnels should also be evaluated. To facilitate the design and planning of modern rail transit systems, research on the passenger flow patterns, crowd behaviors, crowding control and crowding level assessment are necessary. Advanced technologies and tools which can support decision makers on designing, operating, and managing rail transit systems are needed. In this thesis, the crowd dynamics-related problems in rail transit systems are studied based on pedestrian flow and evacuation modeling techniques with the help of necessary field studies. The passenger flow phenomena in metro stations and the emergency evacuation process from rail tunnels are the focuses while the methodologies introduced here are also applicable to many other similar contexts. An agent-based continuous microscopic pedestrian flow simulation model is developed for this purpose. Locomotion of the agent is realized by using a utility maximization approach for deciding the pedestrian movement direction at each step. New utility functions, movement speed calculating mechanism, and collision avoidance methods are proposed to approach the pedestrian movement behaviors more realistically. The model is calibrated and validated by doing parameter analyses and comparing with published empirical results. Simulation results indicate that the model is able to reproduce the well-known pedestrian lane formation phenomenon. The simulation results of fundamental density-speed relationship and the bottleneck flow rate agree with the empirical results well. The model is also compared with two typical discrete models in terms of performances on simulating normal pedestrian flows and emergency evacuations. A modified Lattice-Gas model and basic floor-field Cellular Automaton model are used for this purpose. The agent-based continuous model is extended for emergency situations by considering several human behaviors in emergencies. These two types of models are basically comparable overall, but discrepancies also exist for some cases. The comparison work is helpful for better understanding the underlying modeling technologies of continuous models and discrete models. The agent-based model is also extended for studying the passenger flow phenomenon in metro stations by considering necessary passenger behaviors. Comprehensive Field surveys on passenger flow in metro stations were carried out in 10 of the busiest metro stations in Hong Kong. Over 25 people were employed to collect the patronage data as well as the passenger behavior data. Crowding assessment methods are proposed based on both the field study results and the passenger flow simulations. These crowding assessment methods are useful for station operators, planners, or designers to carry out both qualitative and quantitative analyses of the passenger flow and congestion level in a station. Another type of crowding problem studied in this research is the evacuation process from rail tunnels in emergencies. The extended agent-based model for emergency situations is used for investigating various factors that may affect the evacuation process. Hypothetical tunnel evacuation scenarios were set up for this purpose. The influences of the dimensions of the tunnel elements, including the walkway width and the cross-passageway spacing, on the evacuation time and pattern were studied by means of simulation. The results indicate detailed differences are caused by a change in these dimensions. Besides, the "Stop-and-Go" wave phenomenon featured as temporarily interrupted and longitudinally unstable crowd flow propagating in the opposite direction of the movement was observed in our simulation. The reasons and the factors that may affect the emergence and propagation of the waves are studied in detail.