Temporal attention aware dual-graph convolution network for air traffic flow prediction

Air traffic flow prediction is vital for its supporting function for collaborative decision making in Air Traffic Management. However, due to the inherent spatial and temporal dependencies of air traffic flow and the irregular sector structure in which flow operates in, it is still a challenging problem. To solve this problem, numerous methods are proposed considering airspace adjacency, while flight routes and the origin-destination dependency are not taken into account. In this paper, we propose a temporal attention aware dual-graph convolution network (TAaDGCN) to predict air traffic flow, in which the airspace structure and routes of flow are both included. Firstly, a complementary spatial dual graph convolution module is constructed to capture the dependencies of adjacent sectors and origin-destination (OD) sectors. Then, to include long path information, a spatial embedding (SE) block is adopted to represent potential related sectors of flight traversal. Furthermore, to characterize temporal evolution pattern, a temporal attention (TA) module is applied to access past features of input sequence. Based on the blocks stated above, a spatio-temporal block is constructed in which multiple spatial and temporal dependencies are covered. The experimental results on real-world flight data demonstrate the proposed method can achieve a better prediction performance than other state-of-the-art comparison methods, especially superior to the methods that ignore the sector spatial structure.