A Deep Learning Approach for Aircraft Trajectory Prediction in Pre-Tactical Stage

Abstract

Four-dimensional trajectory prediction is the key technology and research focus to realize the intelligent control of the new generation of air traffic. However, in the complex airspace environment, how to ensure the accuracy and robustness of the trajectory prediction model is the focus and difficulty of the current trajectory prediction research. Aiming at the long-term 4D trajectory prediction in pre-tactical stage, this paper extracts the flight characteristics and typical trajectory change points of the global route. Based on the meteorological information, an aircraft intention prediction method based on Bayesian theory is proposed. On this basis, a probability trajectory generation model based on mixture density LSTM network is proposed. The predicted trajectory information is generated by probability distribution function, and verified by real historical data. The results show that the meteorological information plays a positive role in aircraft intention prediction. The average prediction error of the long-term trajectory prediction model proposed in this paper is less than 1km, which has better prediction accuracy and robustness than common models.