基于模糊kohonen聚类网络的改进算法

Improved Clustering Algorithm Based on Fuzzy Kohonen Clustering Network

模糊kohonen聚类神经网络将模糊隶属度概念应用于一般Kohonen聚类网络的学习和更新策略中,改善了Kohonen聚类网络的性能,是一种快速有效的聚类网络。但在死神经元的处理和收敛速度上还有改进的空间。为了能使网络更好应用于海量数据的聚类问题,对模糊Kohonen聚类网络算法在输出神经元的模糊偏置度、侧抑制模糊隶属度和加权系数提出了三方面改进。同时,对改进的模糊Kohonen聚类网络的有效性进行实例仿真,仿真结果体现了改进算法能有效避免死神经元的出现和提高了网络的聚类速度。

As a fast and effective clustering neural network, Fuzzy Kohonen Clustering Network applies the concept of fuzzy membership degree to Kohonen＇s study and upgrading strategies, which improves the function of Kohohen Clustering Network, yet is to be improved when dealing with dead neurons and convergence rate. For a better application of the network in clustering mass data, clustering algorithm of FKCN should be improved in the aspects of output nerve unit of fuzzy bias degree, lateral inhibition of fuzzy membership and weighting coefficient. Meanwhile, the simulation about the effectiveness of clustering algorithm has been carried out. The results have showed that the improved algorithm can avoid dead neuron and accelerate clustering rate.