神经网络NARX压电陶瓷执行器迟滞建模  

Neural NARX Model for Hysteresis in Piezoelectric Actuators

作　　者：张新良[1] 贾丽杰 付陈琳 ZHANG Xin-liang;JIA Li-jie;FU Chen-lin(School of Electrical Engineering and Automation, Henan Polytechnic University, Jiaozuo 454000, China;Department ofInformation, Beijing University of Technology, Beijing 100022, China)

机构地区：[1]河南理工大学电气工程与自动化学院,河南焦作454000 [2]北京工业大学信息学部,北京100022

出　　处：《控制工程》2019年第5期806-811,共6页Control Engineering of China

基　　金：教育部高校博士点基金-新教师类(20134116120003);国家自然科学基金(U1404612);河南省控制工程重点学科开放实验室基金(KG2016-15)

摘　　要：为了准确描述压电陶瓷执行器中存在的速率依赖性迟滞,给出了一种由迟滞环节和动态环节构成的串联块模型。迟滞环节的数学模型通过引入描述梯度变化的迟滞算子,将多值映射转化为扩展输入空间上的一一映射,使用前向神经网络来逼近。基于迟滞模型,引入线性动态模型描述压电陶瓷迟滞输出的速率依赖性,构造神经网络NARX(nonlinear autoregressive model with exogenous inputs)模型。进一步,给出了模型参数优化的递推随机牛顿算法。实验结果验证了所提出的模型和估计算法的有效性。To describe the rate-dependent hysteresis behavior of the piezoelectric actuators, a cascade block-based model is proposed in this paper, i.e., a rate-independent hysteresis block cascading with a rate-dependent block. For the approximation of the hysteresis block, a hysteresis operator is introduced into the input space to represent the changing tendency of the gradient with the hysteresis. Then a neural hysteresis sub-model is constructed based on a one-to-one mapping. Meanwhile, to describe the rate-dependent characteristics of the dynamic hysteresis, a NARX(nonlinear autoregressive model with exogenous inputs)model is adopted. And a recursive stochastic Newton approximation algorithm is derived for the optimization of the model. The validation results have shown the effectiveness of the proposed model for characterizing the dynamic hysteresis.

关 键 词：压电陶瓷执行器 动态迟滞 神经网络 NARX 随机牛顿算法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]