With the rapid development of society and economy, as well as the fast urbanization, the number of city motors and traffic flow increases dramatically. As a result, the city becomes crowded and congested, lagging the economic development and urbanization process. How to solve the traffic congestion is an important problem the metropolis facing now. Improving the existing traffic signal control system to increase the road capacity is recognized as an effective measure. Urban traffic signal control system is a giant complex system, with strong randomness and nonlinearity. Therefore, the study of urban traffic signal control has great significance in both theory and practice. This dissertation studies the issue of urban traffic signal control. Several traffic signal control methods are developed based on adaptive dynamic programming (ADP), which is a newly proposed intelligent optimization control method. Combining the idea of reinforcement learning, a reinforcement training algorithm is proposed. These methods are applied in urban traffic signal control system to address the coordination problems between intersections. To verify the reliability and practicality of the proposed methods, a series of experiments are conducted on a standard microscopic traffic simulation software. The main aspects of the dissertation are as follows. Firstly, the backgrounds and significance of this research, the purpose and our tasks are introduced. A brief introduction to the research status of traffic signal control is given. The major contents and the organization of the dissertation are also presented. Secondly, a brief analysis of traffic signal control problem, and related intelligent control theories and methods are described. Thirdly, several optimal traffic signal control algorithms based on the ADP and Q-learningare presented. In the simulation platform, the proposed algorithms are tested on a isolated intersection and dual intersections, to verify their validity and reliability. Fourthly, to achieve coordinately optimal control between multi-intersections, a distributed control strategy is adopted and a reinforcement training algorithm is proposed. Then the proposed algorithms are tested under various types of road networks and traffic conditions on the simulation platform. Finally, the obtained results are summarized and the future work is addressed.
修改评论