Accurate prediction of traffic flow is of great significance for safeguarding public safety and solving traffic congestion, and plays an important role in urban traffic planning, traffic management, and traffic control. Traffic forecasting is one of the challenging topics in recent years because it is restricted to urban road networks and changes with time, and there are spatial dependence and time dependence. In order to capture both spatial and temporal dependencies, a new neural network is proposed: A space-time map convolutional network based on the attention mechanism (A-TGCN). The TGCN network model is used to capture the dynamic spatiotemporal characteristics and correlations in traffic data, and an attention mechanism is used to enhance the information of key nodes in each A-TGCN layers. The experimental results on two sets of data show that A-TGCN has a good performance in terms of accuracy and interpretability.