摘要
超高速跨域飞行、敏捷机动等是新一代飞行器发展方向,而长时高速飞行产生的气动外形变化带来的气动参数大范围改变等问题,都对控制系统设计提出了更高的要求。为提高飞行器对模型不确定性的适应能力及控制方法对不同外形、复合执行机构的通用性,深入研究了弱模型依赖的通用智能姿态控制技术,分层次地开展了基于深度学习(DL)的自适应姿态控制、基于深度确定性策略梯度算法(DDPG)的通用姿态控制、弱模型依赖的多维复合控制等技术研究,显著提高了控制系统的鲁棒性和通用性,对人工智能技术在飞行器姿态控制中的应用具有一定的指导意义。
Ultra-high speedcross-domain flight and agile maneuvering are the developing trends of next-generation aircrafts. However,the aerodynamic parameters variations caused by the aerodynamic shape change in long-time widespeed-range hypersonic flight and the aerodynamic variations in deformable aerial-underwater flight pose significant challenges to the aircraft attitude control system. In this paper,a novel generalized intelligent attitude control method with weak model dependence is proposed to tackle the model uncertainty as well as the compound control problem of heterogeneous actuators in deformable aircrafts. The method is an attitude control scheme based on an adaptive control method,a generalized intelligent attitude control method,and a compound control method. The adaptive attitude control method is based on deep learning(DL),and is used to compensate the aerodynamic moment. The generalized intelligent attitude control method is based on the deep deterministic policy gradient(DDPG) algorithm,and is developed for the aerodynamic and model uncertainties. The compound control method is adopted for the heterogeneous actuators with weak model dependence. The proposed method is a practical intelligent control method,and has better robustness as well as universality compared with the existing ones.
作者
邵会兵
詹韬
付京博
SHAO Huibing;ZHAN Tao;FU Jingbo(Beijing Institute of Control and Electronic Technology,Beijing 100038,China)
出处
《上海航天（中英文）》
CSCD
2022年第4期66-75,共10页
Aerospace Shanghai（Chinese&English）
关键词
弱模型依赖
自适应智能控制
多维复合控制
深度强化学习(DRL)
扩张状态观测器
weak model dependence
adaptive intelligent control
heterogeneous compound control
deep reinforcement learning(DRL)
extended state observer
