基于对角回归神经网络的自整定全系数自适应控制器及其特征参量辨识  

DRNN Based Self-Adjusting of All-Coefficients Adaptive Controller and Identification of Its Characteristics Parameters

作　　者：王丽娇[1,2] 

机构地区：[1]北京控制工程研究所,北京100190 [2]空间智能控制技术重点实验室,北京100190

出　　处：《航天控制》2011年第5期15-21,共7页Aerospace Control

摘　　要：针对全系数自适应控制器参数人为调试带来的不便,提出一种基于对角回归神经网络的参数自整定方法,通过神经网络的自学习能力对全系数自适应控制器参数进行在线整定。同时,提出一种新的特征模型参数间接辨识方法,采用神经网络的权值和回归层输出组成的非线性函数构造对象的特征参量,更有效地对特征模型的时变参数进行自学习和调整。对闭环回路的仿真结果表明,基于神经网络辨识和整定的全系数自适应控制回路比传统的全系数自适应控制回路具有更强的自适应性和抗干扰能力,响应速度更快,系统鲁棒性更强。A self-adjusting method based on DRNN is proposed to solve the inconvenience of the adjusting of all-coefficients adaptive controller. The controller can change its parameters on line automatically for the self-learning ability of the neural network. In addition, a new identification algorithm of the characteristics model parameters is presented. In the algorithm, the nonlinear function derived from the weight and output of the recurrent layer of DRNN is used to establish the characteristics parameters. The characteristics parameters can be adjusted more effectively by using the neural network. The simulation of the closed loop shows that the proposed method has better robustness ,performance in response, adaptability and resistance to disturb. The robustness of the system is improved.

关 键 词：全系数自适应控制器 对角回归神经网络 自整定 特征模型 辨识 

分 类 号：V418[航空宇航科学与技术—航空宇航推进理论与工程]