Abstract: A new intelligent urban traffic real-time control system has been designed and completed by the authors and their colleagues. This paper analyses its basic principle from the point of view of transportation engineering, computer science, system structure and control method, etc., The schematic diagram of the system including the main control loop and the machine learning loop is given. Its inference process (traffic flow state inference process and real-time control variable inference process) and machine-learning rules are described. The treatment principle of cycle, offset and split is introduced.