基于自组织聚类和JS散度的RBF神经网络
RBF neural network based on self-organizing clustering and JS divergence
董镇林 1伍世虔 1叶健 1银开州2
作者信息
- 1. 武汉科技大学信息科学与工程学院,湖北武汉 430081;武汉科技大学机器人与智能系统研究院,湖北武汉 430081
- 2. 北京奥信化工科技发展有限责任公司项目部,北京 100040
- 折叠
摘要
针对如何确定径向基函数(RBF)神经网络隐层结构这一问题进行研究,提出一种基于自组织聚类和JS散度的RBF神经网络.为解决K-means算法对初始值敏感的问题,提出基于距离的自组织初始聚类,将戴维森堡丁(DBI)指数作为准则函数,进一步提高聚类精度,得到代表数据集分布特性的隐节点;为解决隐节点冗余和相似的问题,提出一种基于敏感度分析的隐节点删除方法和基于詹森-香农(JS)散度的隐节点合并方法.仿真结果验证了该算法的有效性.
Abstract
Aiming at the problem that how to determine the hidden layer structure of radial basis function(RBF)neural network,an RBF neural network based on self-organizing clustering and JS divergence was proposed.A distance-based initial clustering was proposed to resolve the problem that the K-means algorithm is sensitive to the initial values.The Davies-Bouldin Index(DBI)was used as the criterion function to further improve the clustering accuracy,and the hidden nodes representing the distri-bution characteristics of the data set were obtained.The hidden nodes deletion method based on sensitivity analysis and the hid-den nodes merging method based on Jensen-Shannon(JS)divergence were proposed to resolve the problem of redundant and similar hidden nodes.Experimental results of the simulation verify the effectiveness of the algorithm.
关键词
RBF神经网络/隐层结构/自组织聚类/K-means算法/戴维森堡丁指数/敏感度分析/詹森-香农散度Key words
RBF neural network/hidden layer structure/self-organizing clustering/K-means clustering/Daveis-Bouldin index/sensitivity analysis/Jensen-Shannon divergence引用本文复制引用
基金项目
国家自然科学基金面上基金项目(61775172)
出版年
2024
NETL
NSTL
万方数据