FT-TF: A 4D Long-Term Flight Trajectory Prediction Method Based on Transformer

The accurate prediction of target aircraft trajectory in the process of air combat can significantly improve the ability of aircraft to gain air superiority. Most of the trajectory prediction methods currently applied in air combat are based on traditional time-series prediction algorithms such as LSTM with short prediction steps, which can not realize the demand for long-term prediction in air combat. To address the issue, a novel prediction method FT-TF is proposed in this paper. FT-TF is a multi-step 4D flight trajectory prediction method based on Transformer applied to further improve the accuracy and practicality of trajectory prediction problems. FT-TF uses the first 30 steps of aircraft historical trajectory data as input and the last 30 steps as output. Flight trajectory data sets of aircraft in air combat are used in the experiment to make the trained model closer to the real air combat scene. The experimental results show that the proposed FT-TF can achieve higher prediction accuracy than LSTM and BP neural network. Related Transformer variants of the algorithm are also added to the experiment for comparison, the results demonstrate that the proposed method achieves the smallest prediction error.

1 Introduction

Trajectory prediction in air combat is a process of reason-ably calculating the possible location of the target aircraft in the future. In the process of air-to-air confrontation under the information battlefield environment, accurate trajectory pre-diction results can provide reliable data support for air com-bat situation assessment, intention recognition and maneuver decision-making. The trajectory prediction of enemy aircraft plays a significant role in forming a continuous OODA[1] cycle faster than the opponent. The excellent trajectory pre-diction ability can enable the attack aircraft to complete the observation process faster and quickly establish advantages in air combat. Therefore, it is of great significance to study the trajectory prediction algorithm of target aircraft for ob-taining air superiority.