【摘要】 针对一般竞争神经网络进行聚类分析时需要事先指定聚类数目 ,并且聚类结果依赖于神经网络初始权值的缺点 ,提出了一种结构自适应的竞争神经网络 ,根据聚类结果的MH(modificationHuber)标记值 ,自适应地调整神经网络输出神经元数目以得到最优的聚类数目。仿真结果表明 ,该网络具有良好的动态聚类效果。更多还原

【Abstract】 A traditional competitive neural network(CNN)has the shortcomings that the output neuron number should be determined in advance for cluster analysis and the clustering result depends heavily on the selection of the initial weights. A new CNN with self adaptive structure is presented, which adjusts the output neuron number according to the modification Huber(MH) mark of the clustering result. The numerical experiment results show that the new CNN can provide excellent cluster analysis results.\;更多还原