基于径向基神经网络的直线超声电机位置控制  

Position Control of Linear Ultrasonic Motor Using Radial Basis Function Neural Network

作　　者：孙志峻[1] 帅双辉 金家楣[3] 姚志远[3] 黄卫清[3] 

机构地区：[1]南京航空航天大学机电学院,南京210016 [2]第二炮兵军代室,南京210013 [3]南京航空航天大学精密驱动研究所,南京210016

出　　处：《振动．测试与诊断》2010年第6期650-653,共4页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金-广东省联合基金资助项目(编号:U0934004);国家自然科学基金重点资助项目(编号:50735002)

摘　　要：针对直线超声电机的精密位置控制,提出了一种基于径向基神经网络的自适应控制机制。鉴于直线超声电机工作原理,其运行状态必然受到摩擦、强非线性和时变等不确定性因素的干扰,为了对这些不确定性因素进行有效的逼近,采用了径向基神经网络。为了提高控制机制的自适应能力,首先利用来自试验数据的训练样本按正交最小二乘算法确定径向基神经网络的隐层单元的个数和相关参数,再按递推最小二乘法在线调整隐层与输出层之间的权重。试验结果表明,基于径向基神经网络的自适应控制器的性能不仅优于传统的PID控制和误差反向传播神经网络控制,而且具有很好的抗干扰能力。This paper presented an adaptive control scheme based on radial basis function neural network to precisely control the position of linear ultrasonic motor.Due to the mechanism of the linear ultrasonic motors,their operation status is subject to friction,strong nonlinear and time-varying uncertainties interference.The radial basis function neural network was adopted to approximate these uncertainties effectively.In order to improve the adaptive preference,firstly,the number of the hidden units and parameters of the network were determined by using the orthogonal least square algorithm with the training data collected from experiments.Then,the weights between the hidden layer and the output layer were updated according to the recursive least square algorithm.The experimental results show the proposed controller is better than PID and error back propagation neural network controllers,and has a good anti-interference capability.

关 键 词：超声直线电机 径向基神经网络 位置控制 最小二乘法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]