基于动态递归神经网络的超磁致伸缩驱动器精密位移控制  

Precision Position Control for Giant Magnetostrictive Actuator Based on Dynamic Recurrent Neural Network

作　　者：曹淑瑛[1] 郑加驹[1] 王博文[1] 黄文美[1] 颜威利[1] 

机构地区：[1]河北工业大学电磁场与电器可靠性省部共建重点实验室,天津市红桥区300130

出　　处：《中国电机工程学报》2006年第3期106-111,共6页Proceedings of the CSEE

基　　金：国家自然科学基金项目(50371025);河北省自然科学基金(503055)~~

摘　　要：由于内在的滞回非线性,超磁致伸缩驱动器(GMA)会在开环系统中引起定位误差,在闭环系统中造成系统不稳定。为了克服这个问题,将动态递归神经网络(DRNN)前馈和PD反馈控制器相结合,提出了一种实时滞回补偿控制策略,以期实现GMA的精密位移跟踪控制。DRNN控制器是根据GMA的滞回特性构造的,通过反馈误差学习方案在线学习GMA的逆滞回模型。仿真结果表明该控制策略能适应GMA滞回特性随机械负载、输入信号的变化,在线建立GMA的滞回逆模型,从而消除滞回非线性的影响,实现GMA的精密控制。Due to the inherent hysteretic nonlinearity, the giant magnetostrictive actuator （GMA） can cause position error in the open-loop systems, and cause instability in the closed-loop systems. To remedy this problem, a real-time hysteretic compensation control strategy combining a dynamic recurrent neural network （DRNN） feedforward controller and a proportional derivative （PD） feedback controller was proposed to implement the precision position tracking control of the GMA. The DRNN controller was constructed based on the hysteretic characteristics of the GMA, and on-line learned the inverse hysteresis model of the GMA by the feedback-error learning scheme. Simulation results show that the proposed control strategy can adapt to the changes of hysteretic characteristics of the GMA under different mechanical loads or input signals, on-line obtain inverse hysteresis model of the GMA, thus eliminate the hysteretic impact and achieve the precision control of the GMA.

关 键 词：超磁致伸缩驱动器 滞回非线性 反馈误差学习 动态递归神经网络 实时补偿控制 

分 类 号：TM153[电气工程—电工理论与新技术]