Multi-Phase Deep Learning for Long-Term Trajectory Prediction of Multi-Stage Rockets

This paper proposes a trajectory prediction method to tackle the challenge of long-term trajectory forecasting for multi-stage rockets with limited observation data. The approach utilizes a multi-phase deep learning architecture to perform sequential tasks, including flight phase identification, engine shutdown time prediction, and control order estimation. By integrating the rocket motion model, the predicted states are iteratively updated to construct the complete flight trajectory. Experimental results demonstrate that the proposed method achieves significantly higher prediction accuracy compared to single-network methods when trained on typical multi-stage rocket trajectory data.