Intelligent Trajectory Prediction Algorithm for Reentry Glide Vehicle via Physics-Informed Constraints and State Predictive Control

Abstract

Traditional parameter estimation-based trajectory prediction algorithms for Reentry Glide Vehicles (RGVs) typically suffer from limitations, as they neglect the influence of state variables and rely heavily on high-dimensional motion information. To address these issues, this paper proposes an intelligent trajectory prediction algorithm for RGV via physics-informed constraints and state predictive control. First, based on linear system theory, we derived an analytical expression for the prediction error in parameter estimation methods and demonstrated the superiority of these methods through simulations. Second, the Transformer network based on parallel generative decoding pioneers a parameter estimation method of state predictive control, effectively enhancing the accuracy of medium-to-long-term trajectory prediction for RGVs and resolving the dependency of such methods on high-precision parameter estimation information. Finally, by incorporating physics-informed loss during network training based on the dynamic constraints between state variables and control inputs, the network is transformed into a knowledge-data dual-driven model. Simulation results demonstrate that compared to traditional parameter estimation methods, the proposed method exhibits higher prediction accuracy and improved robustness, offering significant engineering application value. Specifically, for prediction durations ranging from 50 to 200 s, this method keeps the average prediction error and the maximum prediction error within 1.4 km and 3.4 km, respectively.