摘要
在神经网络在线学习控制中 ,实时性和控制精度是非常重要的两大指标。提出的一类具有多维存储结构的CMAC网络 ,提高了网络的泛化能力和学习速度。利用这一网络 ,针对不确定性机器人系统 ,考虑其标称模型 ,提出了一种新的实时智能补偿控制策略 ,并利用Lyapunov方法得出了系统全局渐近稳定的充分条件和网络学习律。在该控制策略中 ,系统的控制输入由两部分组成 :基于标称模型的计算力矩及补偿输入 ,其中补偿输入为系统标称惯性矩阵与神经网络输出的乘积。
In neural networks based on line learning control, both the reality and the control precision are vital. A new CMAC neural networks with multi dimension memory architecture in which the generalization and learning speed are improved greatly, has been presented. Based on this neural networks, and taking the normal model into account, a new real intelligent compensation control scheme is presented for the uncertain robotic systems. And using the Lyapunov theory, a new sufficient condition for the stability of the uncertain robotic systems and a training algorithm for neural networks are provided. In this control scheme, the control input consisted of two parts: normal model based computed torque input and compensation input which is the multiple of the normal inertia matrix and the output of the neural networks. At last, a simulation example is given to illustrate the effectiveness of presented methods.\;
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2003年第6期738-741,共4页
Systems Engineering and Electronics
基金
国家自然科学基金 ( 69975 0 0 3 )
湖南省自然科学基金 ( 98JJY2 0 4)
关键词
机器人控制
智能控制
补偿控制
神经网络
Robot control
Intelligent control
Compensation control
Neural networks
