北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2004, Vol. 27 ›› Issue (4): 87-91.

• • 上一篇    下一篇

一种改进的神经网络训练算法

王波涛1,2, 吴伟陵2, 吴善培2   

  1. 1. 北京首信股份有限公司, 北京 100016;
    2. 北京邮电大学 信息工程学院, 北京 100876
  • 收稿日期:2003-09-23 出版日期:2004-04-28
  • 作者简介:王波涛(1969—), 男, 博士后。 E-mail:bupttwang@263.net;吴伟陵(1938—), 男, 教授, 博士生导师。 E-mail:weilwu@bupt.edu.cn

An Improved Training Algorithm for Artificial Neural Networks

WANG Bo-tao1,2, WU Wei-ling2, WU Shan-pei2   

  1. 1. Beijing Capitel Corporation Limited, Beijing 100016, China;
    2. Information Engineering School, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2003-09-23 Online:2004-04-28

摘要: 针对递归最小二乘反向传播训练(RLS-BP) 算法在推导过程中因引入近似公式而影响了收敛速度的进一步提高的情况,提出了一种改进的RLS-BP训练算法,它通过修改误差性能测度使推导过程中不引入近似公式,进一步提高了收敛速度。 实验也表明改进的RLS-BP算法比原算法的收敛速度一般要快,有较好的实用价值。

关键词: 模式识别, 神经网络, 训练算法

Abstract: In order to resolve the convergence speed problem due to the approximate formula in the process of the recursive least squares back propagation training algorithm(namely RLS-BP), we propose here an improved RLS-BP algorithm, which is deduced through an un-approximation formula. Experiments show that theconvergence of the improved algorithm is faster than before. The improved RLS-BP algorithm is more effective.

Key words: pattern recognition, neural network, training algorithm

中图分类号: