基于SOM的多维信息可视化研究

Visualization of Multi-dimensional Information Based on SOM

处理数量巨大且富含多维信息的数据正成为人们关注的热点,可视化技术为分析和处理海量信息提供了新的手段.SOM能够通过自组织的方式实现高维数据的降维.本文介绍了基于SOM法的多维信息可视化技术.该方法先通过SOM矢量量化法建立原型矢量,然后通过SOM的批量训练实现降维,并用多维比例缩放法实现降维后的数据在低维空间坐标上的投影.在此基础上,分别阐述了地图网格映射、基于聚类的可视化和基于变量的可视化3种可视化方法.论文最后用SOM方法对2003年中国非典的多维信息进行了可视化分析,验证了该方法的实用性.

Process massive and multi-dimensional information is a research hot in recent years and visualization technology provides a new approach for us to analyze and understand massive information.SOM is a method that can reduce the dimension of high dimension data thorough self organizing.This paper introduced the principle of multidimensional information visualization based on SOM.The method obtains prototype vectors by vector quantization and trains the network by batch training.Then,it projects the trained data into lower dimensional space by multi-dimensional scaling method.It brings out 3 kinds of visualization modes based on SOM which are map grid match,cluster based and variable based.This paper used SOM method to visualize SARS information in China in 2003 and results showed that it was a practical method to analyze multi-dimensional information.