Focusing on the congestion and safety issues of urban rail transit, this paper takes the Wangfuzhuang Transfer Station of Jinan Metro as the object and, based on the social force model, uses AnyLogic to construct a two-level passenger flow simulation model for the station hall and platform. The parameters are derived from research and literature. The size and speed of pedestrians, the capacity of stairs/escalators, the service time of turnstiles and ticket vending/security checks, etc. are set to construct the flow lines for entering and leaving the station and transfer as well as the train arrival and departure rhythm. The passenger flow density and the average transfer time are selected as the evaluation indicators. The simulation reveals that the north and south stairs and escalator entrances of the concourse are the main bottlenecks. The peak density of the platform stairs reaches 4.8 and 4.0 people per square meter respectively, and the average transfer time is approximately 448 seconds. Three types of optimizations are proposed: improving the guidance signs and on-site guidance, optimizing the flow lines to reduce intersections, and coordinating the time of Line 1/2. After the comprehensive implementation, the density of the platform stairs was reduced to 3.4 and 2.5 people per square meter, and the average transfer time was reduced to approximately 334.9 seconds, significantly enhancing the transfer efficiency and safety margin. The research has formed a methodological framework of "investigation - modeling - evaluation - optimization - verification", providing quantitative basis for the passenger flow organization and renovation of similar transfer stations. It also points out that the introduction of the Internet of Things and big data can further improve the parameter accuracy and model applicability