Multiple Aircraft Trajectory Prediction Based on Bayesian Social LSTM

Trajectory prediction is critical for aircraft safety. However, the interactions between aircraft in dense airspace complicate accurate trajectory prediction. To address this challenge, we propose a new probabilistic multiple aircraft trajectory prediction method, Bayesian Social Long Short Term Memory (BS-LSTM), which is designed to capture the spatiotemporal relationships among multiple aircraft under uncertain conditions. First, this paper presents a brief overview of existing trajectory prediction methods, highlighting their limitations in dense airspace scenarios. Next, LSTM networks are constructed for each aircraft, with hidden states shared among neighboring aircraft to enhance the model’s awareness of surrounding conditions. Finally, Bayesian Neural Networks are used to compute the prediction interval of trajectories, providing a probabilistic measure of prediction accuracy. Numerical examples demonstrate the effectiveness of the proposed method, and sensitivity analysis is performed to examine the impact of the prediction/observation length ratio on the model’s performance. The results indicate that the BS-LSTM model significantly improves trajectory prediction accuracy in dense airspace, showcasing its potential for enhancing air-traffic safety.