English

应用科学学报 ›› 2002, Vol. 20 ›› Issue (1): 99-103.

• 论文 • 上一篇    下一篇

基于RBFNN和GA的重叠峰分辨新技术

李一波, 黄小原   

  1. 东北大学工商管理学院, 辽宁沈阳 110006
  • 收稿日期:2000-10-29 修回日期:2001-06-02 出版日期:2002-03-31 发布日期:2002-03-31
  • 作者简介:李一波(1963-),男,吉林伊通人,教授;黄小原(1947-),男,河南罗山人,教授,博导.
  • 基金资助:
    辽宁省教育厅科学基金资助项目(20182241);辽宁省自然科学基金资助项目(972147)

A New Technology Based on RBFNN and GA for Resolving Overlapped Peaks

LI Yi-bo, HUANG Xiao-yuan   

  1. School of Business Administration, Northeast University, Shenyang 110006, China
  • Received:2000-10-29 Revised:2001-06-02 Online:2002-03-31 Published:2002-03-31

PDF

114

被引次数

22

摘要： 根据光谱线型函数和色谱峰型函数的(近似)径向对称性和紧支性,首先构造了以线型或峰型函数为基函数的径向基函数神经网络(RBFNN),在RBFNN学习算法中引入了基于可行域约束和共享小生境技术的遗传算法(GA),从而使RBFNN具有了结构自学习和参数优化的能力.最后将具有结构自学习能力的RBFNN成功地引入至谱图的重叠峰解析辨识研究中,试图建立一种适应光谱和色谱谱图重叠峰解析辨识的统一架构,并达到了预期的目的.为了提高解析辨识的成功率,避免遗传算法的盲目搜索,文章还将参数的约束关系作为罚函数引入至遗传算法的适应值函数中,极大地限制了解的空间,减少了病态解发生的概率.

关键词: 径向基函数神经网络, 遗传算法, 重叠峰, 谱图

Abstract: Radial basis function neural network (RBFNN) based on chromatographic peakshape functions and spectral lineshape functions has been constructed. A genetic algorithm based on feasible field constraint and shared niche technique is introduced to enable this RBFNN to acquire the ability of reconstruction. Next the RBFNN was introduced into the resolution of overlapped peaks of spectrum. This paper tries to build up a integrative frame suited to the resolution of spectrum and chromatogram, and partial results have been obtained.

Key words: RBF neural network, genetic algorithm, spectrum, overlapped

中图分类号: