An Intelligent Trajectory Prediction Algorithm for Reentry Glide Vehicles Based on Physics‐Informed Constraints and Prediction Error Compensation

Le résumé fourni par la source

To address the low accuracy of medium‐ to long‐term (50–200 s) trajectory prediction for reentry glide vehicles (RGVs), this paper proposes an intelligent trajectory prediction algorithm based on physics‐informed constraints and prediction‐error compensation. First, an encoder–decoder architecture leveraging a Long Short‐Term Memory (LSTM) recurrent neural network is designed to achieve full‐segment prediction for both tracking and forecasting sequences, providing a data‐driven baseline for subsequent error modeling. Second, a six‐dimensional physical loss—incorporating altitude, longitude, latitude, velocity, flight path angle, and heading angle—is integrated into the training process to enforce physical consistency, thereby enhancing model robustness and generalization. Finally, by analyzing the residuals between predicted and filtered positions during the tracking phase, four error compensation models combining trend and periodic components are constructed. The optimal model is selected using the Akaike Information Criterion (AIC) to correct the position predictions in the forecast horizon. Simulation results show that the proposed algorithm improves prediction accuracy by over 68% compared with mainstream methods, with both average and maximum errors kept within 5 km.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.