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

作　　者：董镇林 伍世虔 叶健 银开州 DONG Zhen-lin;WU Shi-qian;YE Jian;YIN Kai-zhou(School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,China;Institute of Robotics and Intelligent Systems,Wuhan University of Science and Technology,Wuhan 430081,China;Project Department,Beijing Auxin Chemical Technology Ltd,Beijing 100040,China)

机构地区：[1]武汉科技大学信息科学与工程学院,湖北武汉430081 [2]武汉科技大学机器人与智能系统研究院,湖北武汉430081 [3]北京奥信化工科技发展有限责任公司项目部,北京100040

出　　处：《计算机工程与设计》2024年第4期1062-1068,共7页Computer Engineering and Design

基　　金：国家自然科学基金面上基金项目(61775172)。

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。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 distribution characteristics of the data set were obtained.The hidden nodes deletion method based on sensitivity analysis and the hidden 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 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]