基于SOM网络的股票聚类分析方法_徐志超.pdf

第29 卷 第9 期 计算机工程与设计 2008 年5 月 Vol . 2 9 N o . 9 Computer Engineering and Design May 2008 基于SOM 网络的股票聚类分析方法 1,2 2 2 徐志超 ， 梁艳春 ， 时小虎 (1. 长春税务学院 网络中心，吉林 长春 130117 ；2. 吉林大学计算机科学与技术学院，吉林 长春 130012) ( ) 摘 要：无监督的自组织映射 SOM 神经网络是用于聚类的主要人工神经网络模型之一。在SOM 网络的基础上改进了网络 中的邻域函数，并将其用于对股票进行分析和选择，得到了令人满意的结果。为了提高解的精度，避免多个输入样本映射到 同一输出节点还提出了禁忌映射的方法。数值模拟表明该模型对于上市公司的聚类结果令人满意，对于股民客观、准确地 选出真正具有投资价值的股票具有指导意义。 关键词：SOM 神经网络; 动态竞争; 聚类; 禁忌映射; 股票分析 中图法分类号：TP 183 文献标识码：A 文章编号：1000-7024 (2008) 09-2426-03 Analysis method of SOM-based of stock clustering 1,2 2 2 XU Zhi-chao , LIANG Yan-chun , SHI Xiao-hu (1. Network Center, Changchun Taxation College, Changchun 130117, China; 2. College of Computer Science and Technology, Jilin University, Changchun 130012, China) Abstract ：Unsupervised self-organizing map network, also called SOM, is one of neural networks which are applied to cluster. The neighborhood function is improved in SOM network and the improved SOM is used to analyze and select stock, and the results are sati- sfied. A tabu-mapping method is proposed to avoid that the same output node is mapped by more than one input. Numerical experiments show that the clustering results of the proposed model for the companies in the stock market are fairly good. Therefore this model might be beneficial for the investors to select stocks obj ectively and correctly and then get more profits. Key words ：SOM neural network; dynami