跨域变构飞行器自学习模型预测姿态控制方法  

Self-learning Model Predictive Attitude Control Method for Large-flight-envelope Morphing Flight Vehicle

作　　者：贾正宇 张冉[1] 李惠峰[1] JIA Zhengyu;ZHANG Ran;LI Huifeng(School of Astronautics,Beihang University,Beijing 102206,China)

机构地区：[1]北京航空航天大学宇航学院,北京102206

出　　处：《宇航学报》2025年第3期499-508,共10页Journal of Astronautics

基　　金：国家自然科学基金(92471204)。

摘　　要：跨域变构飞行器可以根据任务环境自主改变气动外形,以最佳的气动性能完成跨空域、跨速域飞行任务。针对跨域变构滑翔段的动力学模型大范围变化,以及变构过程引起的气动扰动非线性不确定性的问题,提出了一种自学习模型预测姿态控制方法。在随机干扰和构型下,依据系统自身的模型与数据进行控制参数的学习,提高控制系统的鲁棒性。该方法在参数化模型预测控制问题的基础上,将模型偏差和构型作为随机变量,通过参数学习降低随机最优控制问题的代价函数,得到变构飞行器模型预测最优控制参数。仿真结果表明,对不同的构型变化任务,所提出的控制方法能够在30%的气动参数偏差下保持较好的控制品质,且相较于未训练参数能够提升姿态跟踪响应速度。The large-flight-envelope morphing flight vehicle can autonomously adjust its aerodynamic shape according to the environment to achieve optimal aerodynamic performance for large-flight-envelope mission.To address the significant changes in the dynamic model caused by morphing and the nonlinear uncertainties in aerodynamic disturbances during the glide phase,a self-learning model predictive attitude control method is proposed.This method involves learning of control parameters based on the system’s own model and data under random disturbances and configurations to enhance the controller's robustness.Based on the parametric model predictive control problem,this method treats model deviations and configurations as random variables to reduce the cost function of the stochastic optimal control problem through parameter learning,and obtains the optimal model predictive control parameters for the morphing flight vehicle.Simulation results for different morphing missions show that this control method can maintain good control quality under a 30%deviation in aerodynamic parameters,and improve attitude tracking response speed compared to untrained parameters.

关 键 词：变构飞行器 参数学习 自学习 模型预测控制 

分 类 号：V448[航空宇航科学与技术—飞行器设计]