Preisach迟滞逆模型的神经网络分类排序

为了补偿影响压电陶瓷执行器纳米定位系统精度的迟滞非线性,提高系统的控制精度,开展了基于压电陶瓷执行器的迟滞非线性逆模型的研究。兼顾到迟滞的擦除特性和建模的精确度,提出了一种Preisach逆模型分类排序法的神经网络实现方法,用神经网络取代了传统的反查值方法,以避免插值误差。建立三层BP神经网络,运用实测数据进行训练,确定各层权值;然后,结合排序得到的电压和位移极值信息,通过神经网络方法拟合出较精确的输入电压值。运用若干组实验数据检验了此逆模型的有效性,结果表明,该神经网络的实现方法将逆模型的平均误差降低到了1.5V以下,最大误差绝对值降低到了2.7V以下。与反查值方法相比,神经网络实现方法有效提高了压电陶瓷执行器纳米定位系统的迟滞逆模型的精度。

Realization of sorting & taxis of Preisach inverse hysteresis model using neural network

In order to compensate the hysteresis nonlinearity and to improve the precision of the nanometer positioning system with hysteresis in a piezo-ceramic actuator, this paper studies the inverse hysteresis modeling of the piezoceramic actuator. Taking both the wiping-out property and the modeling precision into consideration, a neural networks is proposed to realize the sorting &; taxis model of hysteresis and to replace the reverse checking and interpolation method to reduce the error of the hysteresis modeling. A BP network with three layers is established,and the weight for every layer is obtained by training practical data. Based on the voltage and displacement extrema got from sorting and taxis, the input voltage of the piezoceramic actuator is obtained by using the neural network.Furthermore, several groups of experiment data are used to verify the accuracy of the proposed inverse model. Results indicate that this method using neural network reduces the average error of the input voltage to less than 1.5 V and the max error of the absolute value to less than 2.7 V. Compared with the reverse checking and interpolation method, this method effectively improves the precision of the Preisach inverse hysteresis model.