计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (5): 96-98.

• 学术探讨 • 上一篇    下一篇

一种基于代数算法的RBF神经网络优化方法

张江涛,刘旭敏   

  1. 首都师范大学 信息工程学院,北京 100037
  • 收稿日期:2007-06-12 修回日期:2007-08-15 出版日期:2008-02-11 发布日期:2008-02-11
  • 通讯作者: 张江涛

Optimization approach based on algebraic algorithm for RBF neural network

ZHANG Jiang-tao,LIU Xu-min   

  1. College of Information Engineering,Capital Normal University,Beijing 100037,China
  • Received:2007-06-12 Revised:2007-08-15 Online:2008-02-11 Published:2008-02-11
  • Contact: ZHANG Jiang-tao

摘要: 提出了一种新的RBF神经网络的训练方法,采用动态K-均值方法对RBF 神经网络的隐层中心值和宽度进行了优化,用代数算法训练隐层和输出层之间的权值。在对非线性函数进行逼近的仿真中,验证了该算法的有效性。

关键词: 径向基函数神经网络, 代数算法, 动态K-均值方法

Abstract: A new training method is presented for RBF neural network.Moving k-means clustering algorithm is used to optimize the centers and widths of RBF algebraic algorithm is used to train the weights between hidden layer and output layer.The approach is used in the approximation of nonlinear function.And the result indicates it’s effective.

Key words: RBF neural network, Genetic algorithm, Moving k-means clustering algorithm