Large-Scale Flight Delay Propagation Prediction Based on Spatio-Temporal Dynamics Network

Capturing the delay propagation mechanism between multiple airports is fundamental to delay prediction, which is crucial for aviation professionals to make correct decisions. The essence of this problem is how to effectively utilize the temporal and spatial characteristics of flight data and other influencing factors. However, previous research has only considered fragmented spatio-temporal features and other influencing factors, or used dynamic models to simulate the delay propagation mechanism. Spatio-Temporal Dynamics Network (STDN) is proposed in this paper, which combines propagation dynamics models with deep learning methods for the prediction of large-scale flight delays. The model adaptively improves the parameter learning method of the Graph Attention Network (GAT), and combines the propagation dynamics method for knowledge representation and reasoning. It predicts the number of flight delays by modeling the flight delay propagation process. Using real historical flight data for prediction, the results show that STDN effectively learns the flight delay propagation pattern.