A Hybrid-Driven Model for Trajectory Prediction of Maneuverable Reentry Vehicle Considering Multiscale Correlation

The reentry vehicle (RV) presents significant challenges to trajectory prediction due to its high speed, maneuverability, and complex flight trajectory. The data-driven trajectory prediction method is difficult to capture the periodic characteristics of RV states and lacks dynamic information, resulting in low prediction accuracy and poor interpretability. In this paper, a hybrid-driven model considering multiscale correlation is proposed for RV trajectory prediction. The model incorporates the RV’s dynamic equation through a hybrid modeling method and captures time-varying correlations between RV states at multiple time scales, which improves both prediction accuracy and generalization ability. First, the state of the RV is decomposed into data-driven and physical components using a hybrid modeling approach to improve the model’s interpretability. Then, addressing the complex correlations between sequences at different time scales, frequency domain analysis and adaptive graph convolution are employed to capture the intersequence space–time dependencies. Additionally, the self-attention mechanism is utilized to extract remote intrasequence dependence information. Finally, a generative decoder is designed to predict the data-driven component, thereby reducing cumulative errors, while the physical component is predicted using dynamic equations to achieve hybrid-driven RV trajectory prediction. Simulation results demonstrate that the proposed model outperforms both the physics-informed neural network and transformer methods in terms of prediction accuracy. And the model shows superior generalization ability when applied to unknown datasets.