基于深度强化学习的变体飞行器智能参数整定  

Intelligent parameter adjusting of morphing aircraft based on deep reinforcement learning

作　　者：郭鸿飞 宁国栋[1] 张科南[1] 高广林 Guo Hongfei;Ning Guodong;Zhang Kenan;Gao Guanglin(Beijing Institute of Mechanical and Electrical Engineering,Beijing 100074,China)

机构地区：[1]北京机电工程研究所,北京100074

出　　处：《空天技术》2024年第5期60-70,80,共12页Aerospace Technology

摘　　要：针对变体飞行器内部气动系数时变导致的控制律设计困难、参数调试复杂的问题,提出了一种基于深度强化学习的智能控制策略。将变体飞行器纵向运动模型分解为姿态子系统与速度子系统。对于姿态子系统,在系统非仿射的情况下,采用反步法设计了鲁棒控制律,并基于深度确定性策略梯度算法进行智能参数整定和控制补偿。对于速度子系统,利用深度确定性策略梯度算法智能整定PI控制器参数,实现对速度指令的跟踪。通过训练,智能算法可以根据当前状态快速且智能地调整控制律参数,并通过仿真验证了智能算法的学习效果与控制律的有效性和鲁棒性。Aiming at the difficulty in designing control law and adjusting control law parameters caused by the time-varying aerodynamic coefficients of morphing aircraft,an intelligent control strategy based on deep deterministic policy gradient is proposed.Decompose the longitudinal motion model of the morphing aircraft into attitude subsystem and velocity subsystem.For the attitude subsystem,a robust control law is designed using the backstepping method in non-affine situation,and the parameters of control law are adjusted intelligently and control instructions are compensated based on deep deterministic policy gradient.For the velocity subsystem,the deep deterministic policy gradient is used to intelligently adjust the PI controller parameters and achieve tracking of velocity command.Through training,intelligent algorithm can quickly and intelligently adjust control law parameters based on the current state.The effectiveness and robustness of the proposed methods are verified,as well as the learning effectiveness of the intlligent algorithm through simulation.

关 键 词：变体飞行器 深度确定性策略梯度 智能参数整定 非仿射系统 反步法 PI控制 

分 类 号：V249.1[航空宇航科学与技术—飞行器设计]