摘要
为了研究具有外部干扰力矩和模型不确定性的多体航天器姿态快速跟踪控制问题,基于逆系统方法和回声状态网络(echo state network,ESN),设计了鲁棒控制器,并利用Lyapunov稳定性理论证明了控制系统的渐近稳定性。采用"meta-learning"策略离线训练ESN网络,并应用遗传算法优化其主要参数,解决了动态递归神经网络训练困难及网络参数不易确定的问题。控制器的设计过程相对简单,不需要精确的动力学模型。该文还针对航天器中心体与天线同时跟踪不同目标的任务进行了数值仿真,结果表明所设计的鲁棒控制器对外部干扰与模型不确定具有很好的鲁棒性。
This paper investigates the attitude tracking control of multi-body spacecraft with external disturbance torques and parameters perturbations. A robust controller was developed for attitude tracking based on the inverse system method and the echo state network （ESN）. The asymptotic stability of the closed-loop control system is guaranteed by Lyapunov stability theory. The linear output weights of the ESN can be trained off-line using the meta-learning strategy to overcome the shortcoming of recurrent networks. The ESN＇s primary parameters are optimized in the genetic algorithm to remove the difficulty of choosing the ESN parameters. This control approach required no prior knowledge about the dynamic model. Simulation results show the good tracking performance of the control scheme during spacecraft inertia changes and in presence of external disturbances.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第8期1362-1367,共6页
Journal of Tsinghua University(Science and Technology)
关键词
多体航天器
动态递归神经网络
姿态跟踪
multi-body spacecraft
recurrent neural networks
attitude tracking
