ISSN 1000-1239 CN 11-1777/TP

• 论文 • 上一篇    下一篇

多获胜节点SOM及其在股票分析中的应用

王丽敏1, 2 梁艳春1 韩旭明1, 3 时小虎1 李 明2   

  1. 1(吉林大学计算机科学与技术学院国家教育部符号计算与知识工程重点实验室 长春 130012) 2(长春税务学院计算机科学与技术系 长春 130117) 3(长春工业大学信息传播工程学院 长春 130012) (ycliang@jlu.edu.cn)
  • 出版日期: 2008-09-15

Multi-Winners SOMs and Their Applications to Stock Analysis

Wang Limin1,2, Liang Yanchun1, Han Xuming1,3, Shi Xiaohu1, and Li Ming2   

  1. 1 (College of Computer Science and Technology, Jilin University, Key Laboratory of Symbol Computation and Knowledge Engineering of the Ministry of Education, Changchun 130012) 2(Department of Computer Science and Technology, Changchun Taxation College, Changchun 130117) 3(Institute of Information Spreading Engineering, Changchun University of Technology, Changchun 130012)
  • Online: 2008-09-15

摘要: 为了增强自组织映射(self-organizing map, SOM)网络的动态竞争和聚类能力,提高解的精度,在无监督的SOM神经网络的基础上,通过拓广获胜节点的数量,改进网络中的邻域函数和连接权函数等方法,提出具有多获胜节点的SOM模型.为了避免多个输入样本映射到同一个输出节点,还提出了禁忌映射的方法.为了验证所提出的方法的有效性,以股票的聚类分析为实例,对该方法进行了检验.通过对每股收益、每股净资产、净资产收益率、每股经营性现金流量及净利润等5项反映上市公司综合盈利能力的财务指标进行了模拟实验,所得的数值结果表明,在标准SOM及所提出的几种多获胜节点SOM网络模型中,具有双获胜节点(SOM with 2 winners, SOM2W)的网络模型获得了最好的聚类效果.结合实验结果对网络模型的进一步分析也表明,SOM2W的聚类能力优于标准SOM及其他网络模型.该模型为股票的分析和选择提供了一种可行的途径,在金融领域具有潜在的应用价值.

关键词: 多获胜节点SOM网络, 动态竞争, 聚类, 禁忌映射, 股票分析

Abstract: In order to enhance the dynamic competition and clustering capability of self-organizing map (SOM) neural network and improve the precision of solutions, multi-winners SOM models are proposed by extending numbers of winners based on an unsupervised SOM neural network and by improving the neighbor function and weight function in the network. In addition, a tabu-mapping method is proposed to avoid that the same output node is mapped by more than one input. The clustering analysis for the stock is used to examine the effectiveness of the proposed models. The financial indexes reflecting the whole performance level of companies are used in the simulated experiments, including earning per share (EPS), net asset per share (NAPS), return on equity (ROE), cash flow per share (CAPS) and net profit. Simulation results show that the clustering effect of the SOM with 2 winners (SOM2W) is the best compared with those from the standard SOM and other proposed multi-winner SOMs. Further analysis for the experimental results also shows that the clustering performance of the SOM2W is superior to the standard SOM and those of other models. The proposed model could provide a feasible approach for analyzing and selecting stock, which has potential applications in the financial field.

Key words: multi-winners SOM network, dynamic competition, clustering, tabu mapping, stock analysis