Abstract: Activity is the source of travel.Temporal-spatial characteristics are important indicators of travelers’ activity features and the causes of congestion.In this paper the authors proposed a multi-agent based reinforcement learning algorithm which could simulate activity time and location choice for travelers.Since road congestion condition was a dynamic parameter in the model,the environment could change dynamically with agent’s behaviors.This reflects interactions among travelers and interactions between travelers and the environment.The result analysis of typical travel patterns,traffic flow distribution and activity location choice show that this algorithm’s simulation could represent actual survey results.The deviation of peak flow is less than 5% while the correlation coefficient of activity location choice is more than 90%.