中国机械工程

• 智能制造 • 上一篇    下一篇

基于LM-BP神经网络的非线性轮廓图优化方法研究

许静;何桢;袁荣   

  1. 天津大学,天津,300072
  • 出版日期:2016-10-25 发布日期:2016-10-21
  • 基金资助:
    国家杰出青年科学基金资助项目(71225006)

A Optimization Method for Non-linear Profile Based LM-BP Neural Networks

Xu Jing;He Zhen;Yuan Rong   

  1. Tianjin University,Tianjin,300072
  • Online:2016-10-25 Published:2016-10-21
  • Supported by:

摘要: 将正交试验设计理论与BP神经网络模型和Levenberg-Marquard算法相结合,提出了一种基于LM-BP神经网络模型的针对输出为非线性轮廓图响应的离线设计优化方法。并结合实例与传统的统计回归建模方法得出的优化结果进行了比较。结果表明基于LM-BP神经网络建模可以避免由于实验误差和试验设计方案所造成的模型系数估计误差,而与标准的BP算法比较,克服了标准BP算法性能不稳定、收敛速度慢、收敛精度低、存在局部最小值等缺点,具有极高的精确性,优化结果令人满意。

关键词: BP神经网络, Levenberg-Marquard算法;试验设计;非线性轮廓图

Abstract: A method to optimize the non-linear profile was presented based on the DOE theory BP neural network model with Levenberg-Marquard algorithm, which were compared with the traditional statistical regression modeling by a example. The results show that the estimation errors of the model coefficients due to the design and experimental errors may be avoided based on LM-BP neural network modeling.Compared with the BP algorithm this method overcomes the standard BP algorithm performance of unstability, slow convergence and low convergence precision, the presence of local minima and other short-comings.The method proposed herein has a high accuracy, optimization results are satisfactory

Key words: BP neural networks, Levenberg-Marquard(LM) algorithm, design of experiment(DOE), non-linear profile

中图分类号: