Research On Urban Traffic Signal Optimal Control Algorithm Based On Approximation Dynamic Programming

With the high speed development of economic and the rapid development of China's automotive industry, the urban traffic congestion phenomenon especially in large city becomes more and more serious; this problem caused many traffic accidents. Therefore, how to effectively relieve traffic congestion, improve the traffic efficiency became the researchers first concern.The traditional way that through expand transportation infrastructure to solve the traffic jam is very limited, another way is through improving the utilization rate of existing traffic to make more reasonable and effective control strategy, In recent years the traffic engineering scholars proposed intelligent traffic control system and did a lot of work around this system. So intelligent traffic control system became the fundamental way of solve urban transportation problem.The previous work is mainly aimed at isolated intersection signal control research work, without considering the regional coordination. With the gradual improvement of modern urban road network, this request when consider traffic control optimization must be from a regional perspective. On this basis,we collect every five minutes traffic data in city intersection as the research object, Combined with the characteristics of traffic flow we puts forward the algorithm based on improved Q learning algorithm and approximate dynamic programming (ADP) traffic signal optimization coordination control, ADP method can real-time change according the system parameters and enhance the system robustness,so it can better adapt to the characteristics of real-time traffic system. This paper adopted average delay time as the control performance evaluation index.Finally, we use the measured data and GLD software to simulation two intersections model, and compared the simulation results with the existing traffic control strategy, the results confirmed the algorithm can effectively reduce the traffic average delay time and is a kind of effective prediction method. On this basis, we also points out some existing problems in this algorithm and the direction of further research.