The application of intelligent transportation system in urban transportation is more and more deeply. To achieve accurate traffic volume prediction has important practical significance for traffic service, traffic engineering planning and traffic risk prevention. Although the traffic volume in the road network has periodicity and correlation, traffic accidents and bad weather will cause the change of the road network structure, and the adjacent nodes show time correlation. In this paper, a graph convolution neural network model of dynamic spatio-temporal based on reachability matrix is proposed from the point of view of dynamic reachability between nodes in traffic network. Based on the convolution network of spatiotemporal graph, the dynamic reachability matrix module of parameter learning is added to express the dynamic and spatio-temporal correlation of road network. The experimental results on the open data set PEMs show that the model can effectively capture the dynamics of the road network, improve the accuracy of traffic volume prediction, and is superior to the existing baseline model.