Aiming at the problem of insufficient characterization of pedestrian dynamic changes in urban flood risk assessment, a multi-agent model based dynamic assessment method for urban flood pedestrian risk is proposed. The initial spatial distribution of pedestrian agents is established based on population heat maps. Urban functional zones are delineated based on points of interest to serve as the base environment layer for pedestrian agent activities. A probabilistic finite state machine and velocity decay curves are combined to characterize pedestrian behavior rules under flood scenarios. A 1D-2D hydrodynamic model is coupled to establish an urban flood multi-agent-based model. Historical rainfall data and population heat maps are used to verify model reliability, and the dynamic changes in pedestrian risk under different rainfall return periods and rainfall occurrence time are assessed. The method is applied to the Huayuan Road area in Jinshui District, Zhengzhou City. The evaluation results show that the relative error between the simulated results of the established model and the actual heat map is less than 15%. When the rainfall return period exceeds 50 a, the number of medium-to-high-risk pedestrian agents increases significantly by 143%. The number of medium-to-high-risk pedestrian agents on weekdays is significantly higher than on weekends due to rigid travel demand by 49%. The number of medium-to-high-risk pedestrian agents peaks during the morning rush hour and evening rush hour. Keywords: urban flood; multi-agent model; pedestrian risk; population heat map; Zhengzhou City 〖FL