基于自组织特征映射神经网络的聚类分析

在深入研究自组织特征映射（Self-organizing Feature Mapping,SOFM）神经网络的结构和聚类算法的基础上,阐述了SOFM网络的建立方法.以随机二维向量的聚类为例,利用所建立的SOFM网络模型对输入的随机二维向量进行聚类,并着重研究了输出层神经元拓扑结构、训练步数对聚类结果的影响以及在相同拓扑结构条件下,SOFM网络模型的权值向量的调整过程.仿真结果表明：在输出层神经元节点形式为六边型条件下,输出层神经元的个数越多,SOFM网络模型的聚类结果就越准确;在相同的拓扑结构条件下,训练步数越大,SOFM网络聚类结果越准确,但过大的训练步数对于聚类结果的影响甚微.

Clustering analysis based on SOFM neural network

In this paper, an establishing method of SOFM network is introduced based on an in-depth study on its structure and clustering method. The clustering of two-dimensional vectors is taken as an example, and the established SOFM network is used to cluster the random two-dimensional vectors which are put in. The effects of topology structure and training steps of output layer neurons upon clustering results are studied. Besides, the adjustment process of weight vectors of SOFM network model with the same topology structures is also under investigation. The simulation result shows that when the neuron nodes in the output layer are in the form of a hexagonal, the more the neurons in the output layer, the more precise the SOFM network model＇ s clustering result becomes; when the topology structures are the same, the more training steps, the more precise the clustering result is. However when the number of training steps are too big, there will be little impact on the clustering result.