助推-滑翔飞行器再入过程强化学习自抗扰控制  

Reinforcement learning-based active disturbance rejection control for boost-glide vehicle in re-entry phase

作　　者：浦甲伦[1] 詹韬[2] 李博皓 薛玉兰 Pu Jialun;Zhan Tao;Li Bohao;Xue Yulan(School of Astronautics,Harbin Institute of Technology,Harbin 150001,China;Beijing Institute of Control and Electronics Technology,Beijing 100049,China)

机构地区：[1]哈尔滨工业大学航天学院,哈尔滨150001 [2]北京控制与电子技术研究所,北京100049

出　　处：《战术导弹技术》2024年第2期117-125,共9页Tactical Missile Technology

基　　金：国家自然科学基金重点项目(52232014);国家自然科学基金联合基金项目(U2241215)。

摘　　要：针对助推-滑翔飞行器再入过程强耦合、快时变、强非线性的特点带来的姿态控制问题,提出了一种基于最大熵强化学习参数整定的线性自抗扰控制方法。利用输入-输出反馈线性化建立了三通道解耦的线性控制模型,并设计了线性自抗扰控制器,获取了待设计控制器参数。建立了控制模型的马尔可夫决策过程,利用最大熵强化学习算法对控制器参数进行寻优。针对飞行器再入滑翔段进行六自由度仿真,仿真结果表明,所设计的控制器能够自适应调整控制器参数,与线性自抗扰控制器相比,跟踪速度更快,误差更小,从而验证了所提方法应用于助推-滑翔飞行器再入段姿态控制问题的有效性。A maximum entropy reinforcement learning(RL)-based active disturbance rejection controller is proposed for the boost-glide vehicle in the re-entry phase under coupling and time-varying dynamics and strong nonlinearity.The control-oriented decoupled model is established by the input-output feedback linearization,and the linear active disturbance rejection controller is designed.The Markov Decision Process for the control model is established,and the RL algorithm is utilized to optimize the control parameters.Various numerical simulation for the re-entry phase of vehicle are carried out.The simulation results show that the designed controller is able to adaptively adjust the controller parameters,track the commands more quickly and reduce the control eror compared with the linear active disturbance rejection controller,demonstrating the effectiveness of the proposed control scheme applied to the boost-glide vehicle.

关 键 词：助推-滑翔飞行器 再入段 姿态控制 非线性动力学 反馈线性化 自抗扰控制 最大熵强化学习 六自由度仿真 

分 类 号：V448.1[航空宇航科学与技术—飞行器设计]