摘要
小脑模型关节控制器(CMAC)是一种结构简单、学习速度快的局部神经网络,适合于实时控制。但对于某些高维输入系统来说,CMAC需要大量的存储空间,实际应用性能下降。文章结合传统PID反馈控制与神经网络逆模控制的特点,利用CMAC网络对机械手进行逆模控制;针对网络所需输入量较多的问题,提出了一种单输入CMAC网络的逆模控制策略,并将提出的控制策略应用于2自由度机械手的轨迹控制;引入测量变量使网络输入由二维转换为一维,减少了神经网络所需存储空间,提高了学习速度。仿真实验结果表明,所提出的控制策略克服了机械手非线性和不确定性的影响,是可行的。
Cerebellar model articulation controller（CMAC）is a simple,fast learning neural network which is suitable for real-time control.However,for some high-dimension inputs systems,CMAC requires a lot of storage space,which reduces the practical usefulness.In this paper,CMAC is utilized to implement robot arm inverse model control in view of the characteristics of traditional PID feedback control and neural network inverse model control.A single-input CMAC inverse model control strategy is proposed to solve high-dimension inputs problem.In simulating a 2-DOF robot arm control,the strategy greatly improves the storage space and the learning speed with the help of signed distance which converts two-dimension inputs to one-dimension input.The simulation results show that the proposed scheme counteracts the disadvantageous influence of nonlinearities and uncertainties in robot arm and performances well.
出处
《合肥工业大学学报（自然科学版）》
CAS
CSCD
北大核心
2015年第4期454-457,共4页
Journal of Hefei University of Technology：Natural Science
关键词
神经网络
小脑模型关节控制器(CMAC)
逆模控制
机械手
轨迹控制
neural network
cerebellar model articulation controller（CMAC）
inverse model control
robot arm
trajectory control
