With the rapid development of aviation, the air traffic flow is increasing exponentially, leading to increasingly serious route congestion and route conflicts. Accurate prediction of aircraft trajectories is essential to ensure the safety of airspace. Trajectory data is a time-series sequence, and most of the existing methods for trajectory prediction are based on recurrent neural networks and attention mechanisms. However, these methods fail to effectively accumulate the correlation between the attributes of trajectory points and the time varying attribute features, and suffer from slow model training. In this paper, we propose STA-STCN, a novel track prediction model based on spatio-temporal dual attention (STA) and stacked temporal convolutional network (STCN). STA extracts the spatial and temporal features in parallel based on self-attention. The spatial feature characterizes the relationships between attributes of a track point, and the temporal feature represents the time varying characteristics of an attribute in track points. STA then fuses the extracted features to describe the accumulated time varying attributes correlation. STCN takes the fused features as the input to predict target track point attributes. We conduct experiments on the flight data provided by global flight data services website VariFight. Experimental results show that the model can reduce the root mean square error, mean absolute error, and track prediction model training time, indicating that model STA-STCN can achieve effective track prediction.
Track Prediction Based on Spatio-Temporal Attention
Lect. Notes Electrical Eng.
China Conference on Command and Control ; 2022 ; Beijing, China August 17, 2022 - August 19, 2022
Proceedings of 2022 10th China Conference on Command and Control ; Chapter : 32 ; 341-352
2022-08-30
12 pages
Article/Chapter (Book)
Electronic Resource
English
Spatio-temporal graph attention networks for traffic prediction
Taylor & Francis Verlag | 2024
|