The Application Study Of Based Q-learning In Traffic Signal Control Of Single Intersection

Nowadays traffic jams occur widely in diffcrent cities around the world. To avoid this phenomenon, it is neccssary to reinforce the traffic control and supervision. In urban traffic, the single intersection signal control is the basis of intersection coordinated signal control and network-level signal control.Therefore, it becomes a hot research topic.The thesis aims to study the application of agent technology in traffic signal control. The following main aspects are involved in the thesis:On the basis of research on the principle of reinforcement learning, the thesis discusses the five key problems when Q-learning algorithm is applied in traffic signal control. It not only solves the problem of discretization of state space needed in Q-learning algorithm, but also constructs a single intersection traffic signal control system based on Q-learning.According to the theory of Q-learning algorithm, the author proposes three actions based on the change of each phase green time: increase current phase green timeΔseconds; remain unchanged; decrease current phase green timeΔseconds. With the application of this program of three actions, each phase constitutes a transformed control action set. Based on this control action set and period change, two reward functions are established for the purpose of minimizing average delay. At the same time, two different Q-learning algorithms of traffic signal control are proposed, which are Q-learning algorithm with fixed cycle and variable cycle. After the design of the algorithm, detailed steps are given.In the thesis, a typical four phase-controlled intersection is taken as an example through Matlab programming. Q-learning algorithm with fixed cycle and variable cycle is used for signal timing. Compared with Webster method, better effect of Q-learning for a single intersection signal control is verified.Finally, the author summarizes his research work and puts forward problems for further study.