摘要
针对常规SOFM(self-organizing feature map)无监督的神经网络,提出了一种改进的自组织特征映射SOFM神经网络算法。在常规SOFM网络数据聚类算法基础上,分析了其在实际应用中存在的不足,对初始权值设定以及邻域范围选择等方面进行了算法的优化和改进,进而提高了SOFM神经网络聚类算法的正确率、收敛速度和实时性,并利用仿真实验进一步对提出的改进算法进行了验证。
SOFM (self-organizing feature map) algorithm is a clustering method that can cluster on non-supervision condition. An improved algorithm based on SOFM neural network clustering was introduced in this paper. It proposed the basic data clus tering theory on SOFM and found problems in applications. The selection method of initial weights and the scope of neighbor hood parameters were improved to increase the correct rate, convergence speed and computational efficiency of data clustering. The improved clustering algorithm is verified by simulation results.
出处
《河北科技大学学报》
CAS
2012年第6期514-518,共5页
Journal of Hebei University of Science and Technology
基金
河北省自然科学基金资助项目(E2009000703)
关键词
自组织特征映射网络
数据聚类
初始权值
邻域范围
self-organizing feature map
data clustering
initial weights
scope of neighborhood