基于多项式拟合的压电陶瓷迟滞神经网络建模

Neural Network Modeling Based on Polynomial Fitting for Hysteresis Behavior of Piezoelectric Actuator

在压电陶瓷致动器优化设计的研究中,针对压电陶瓷的迟滞非线性特性,提出了一种基于多项式拟合算法的神经网络建模方法。由于压电陶瓷驱动器的迟滞现象是一种多对多的映射关系,而传统的建模方法只能对一对一映射进行建模。为解决上述问题,在对压电陶瓷迟滞现象的形成原因和特点进行深入分析的基础上,采用多项式拟合和神经网络相结合的方法对压电陶瓷驱动器的迟滞现象进行建模。仿真结果表明,采用多项式拟合算法的神经网络建模克服了传统建模方法只能对迟滞曲线进行分段建模的局限性,且拟合精度比较高,神经网络正模型的拟合误差为1.45%,神经网络逆模型的拟合误差为1.16%。表明上述神经网络模型精确地反映了压电陶瓷的迟滞特性。

This paper proposes a neural network modeling method based on polynomial fitting algorithm for nonlinearity and hysteresis characteristics of piezoelectric ceramic. The hysteresis phenomenon of piezoelectric actuator is a many to many mapping relationship, however, it is modeled as one to one mapping in traditional modeling methods. For this reason, we firstly analyze the causes and characteristics of the hysteresis of piezoelectric ceramic, then we combine the polynomial fitting with neural network to model the hysteresis phenomenon of piezoelectric actuator. Sim- ulation results show that the fitting error of neural network model is 1.45% , and the fitting error of neural network inverse model is 1.16%. This indicates that the neural network model reflects the hysteresis characteristics of piezoelectric ceramics accurately.