Regional Traffic Control Method Based On Reinforcement Learning

In recent years, along with development of economy and society, traffic jams are more and more serious. Practice has proved that scheduling when all comes to all is very useful for area traffic management. But, traditional methods of modeling and control are inefficient because area traffic system is nonlinear, fuzzy and uncertain. In order to avoid modeling, reinforcement learning method is applied to area traffic control in this thesis. Its main contents are the following three aspects:1. The basic theories of area traffic control and reinforcement learning are introduced, and the current applications of these theories are analyzed. Base on result of analysis the reason of choosing such a method in this thesis is given.2. A new optimizing method for area traffic control is proposed. This method applies reinforcement learning to optimize cycle of the whole area, and divide the area into several trunks according to their importance, then optimize offsets and splits in trunks' orders.3. Traffic simulation technology and TSIS simulation software are introduced. The framework of runtime extension (RTE) program for TSIS is presented, too. Finally the method proposed in this thesis is carried out in a RTE program. TSIS simulation shows that this method can effectively lessen delay time and heighten average speed.