基于距离和密度聚类的RBF网络结构设计

为确定径向基函数RBF(radial basis function)神经网络隐含层结构,并针对基于距离或密度聚类的RBF神经网络的限制,提出一种基于距离和密度聚类(GDD)算法的RBF神经网络。GDD算法通过计算每个样本的密度,各样本间的距离及相似条件(密度标准、距离标准),相似条件是根据样本分布而改变的,进行样本空间划分,实现无需先验知识及参数,就可以精确识别任意形状的聚类。采用GDD算法聚类,RBF神经网络能确定合适的隐含层节点个数及径向基函数中心。对典型非线性函数逼近及UCI机器学习库实例样本进行实验,结果表明基于GDD算法设计的RBF神经网络具有良好的逼近能力和分类效果,且网络结构更加紧凑。

Structure Design for RBF Neural Network Based on Distance and Density Clustering

To determine the hidden layer structure in radial basis function(RBF)neural network and solve the problem of RBF neural network based on distance or density clustering,RBF neural network based on distance and density clustering(GDD)algorithm is proposed in this paper.The GDD algorithm can divide the sample space and spot any-shape clusters without prior knowledge and parameters based on the density of each sample,the distance between samples and the similar condition(density standard and distance standard),which is changed according to the sample distribution.The GDD-RBF neural network can determine the appropriate number of hidden layer nodes and the radial basis function center.The results indicate that the proposed GDD-RBF neural network has good approximation and classification ability and the network structure is more compact.