精密定位工作平台的神经网络辨识

Identification of Precision Positioning Stage Based on Neural Networks

研制的精密运动平台同时具有死区与迟滞的特性,普通的辨识算法难以有效处理。本文设计了一种具有特殊激励函数和结构的神经网络,可以同时描述精密运动平台的非光滑死区和迟滞特性,并引入广义梯度的概念来改进L-M训练算法,实现对所构建神经网络的建模训练。实际辨识结果表明所提出的辨识方法取得了令人满意的结果。

Due to the non-smooth nonlinear dead-zone and hysteresis, it is difficult to use the traditional identification method to identify the model of the precision positioning stage. In this paper, a special neural network with non-smooth activation function is adopted to model the non-smooth nonlinear dead zone coupled with hysteresis in the stage. Moreover, the generalized gradient is introduced into the Levenberg- Marquardt algorithm so that the non- smooth neural network can be trained. The experimental results are presented to illustrate the performance of the proposed identification scheme.