基于PAM聚类方法的RBF神经网络设计

Design of RBF Network Based on PAM Clustering Algorithm

RBF神经网络中心点的选择对RBF网络性能的好坏至关重要。提出了一种利用PAM聚类算法确定径向基函数的数目、基函数中心及宽度,输出层权值由线性方程组确定的RBF神经网络设计方法。通过对UCI数据集的分类仿真实验,结果表明利用该方法设计的RBF网络具有良好的分类效果,能有效减少网络的训练时间且对孤立点数据不敏感。

In the RBF Neural Network,the choice of the center of Redial Basis Function determines the performance of the RBF Neural Network.In the paper,we propose a method of design RBF Neural Network using the PAM Clustering algorithm to determine the number and center and spread of Redial Basis Function.The simulate experiment of function approximation explains the RBF Neural Network has the performance such as higher accurate classifying and reduction of network train time and not prune to the noise data.