Study of intersection traffic control based on agent

using reinforcement learning technology and combining with the structure of modular network, a new model of Agent controllable structure based on a intersection is proposed. On the basis of the idea of divide and rule methodology, the complex problem of traffic signal control is divided into local expert unit, which improves the efficiency of learning, and the integrated unite will coordinate the outputs of local expert unit, then a controllable signal of the whole system will be gained. Combined it with phase selected unit, the task of traffic signal control is achieved together. Finally, the effectiveness and reliability of the approach is fully proved by PARAMICS simulation.