The urban arterial road is the aorta of the city and plays an important role in increasing the traffic capacity of the road network. Due to high practical risk, the studies of arterial coordinated control are limited. Parallel Transportation System (PTS) offers an effective approach to investigate optimal control method for arterial coordinated control. In this work, the deep Q network is introduced to PTS platform. We proposed a dynamic arterial coordinated control algorithm. All intersections on the arterial road are handled as a whole. The status characteristics of various intersections in an arterial road are extracted by using the deep neural network. Q-learning is used to accomplish decision-making for traffic signal control. Thus, this algorithm can realize optimal control of time-variant traffic flow. We further investigate experimentally the effect of the deep Q network on arterial coordinated control performances, in which the different number of convolution layers and optimizer are adopted respectively. The simulation results show that in the condition of near saturation and initial queue, our algorithm has much lower average vehicle delay and less average number of stops than the typical arterial coordinated control method.