When the building boom of smart city set off all over the world, in order to seize the commanding heights of the construction of smart city, many of the cities in our country have introduced a variety of measures to build smart city planning. Presently, the construction of the smart city in China has stepped into the fast lane and is becoming a nationwide boom. Meanwhile, with the continuous progress of urbanization in China, the number of car ownership has been rapidly increasing and traffic congestion has become a frequent occurrence. In consideration of the nonlinear and time-varying characteristics of traffic, it is difficult for traditional traffic congestion evacuation strategies to cope with the complex changes of the traffic flow due to their poor adaptability. In addition, traditional evacuation assignment strategies are seldom made from the perspective of traffic flow relevance, which results in low efficiency on traffic congestion evacuation. However, during the construction of smart city, the wide application of technologies related to Internet of Things in various fields of smart city, which makes it possible to realize the instantaneity of information perception and technique collection, and helps to improve the sensory ability of the system. Based on this, and under the background of smart city, this paper puts forward a multi-agent-based concept of “model prediction, dynamic calculations, online feedback, beforehand evacuation” and researches on the evacuation strategies of urban traffic congestion.