首页 | 本学科首页   官方微博 | 高级检索  
     

复杂模式保留拓扑的平面映射及其应用
引用本文:颜学峰,陈德钊,胡上序.复杂模式保留拓扑的平面映射及其应用[J].浙江大学学报(自然科学版 ),2001,35(5):529-533.
作者姓名:颜学峰  陈德钊  胡上序
作者单位:颜学峰(浙江大学,化学工程学系,浙江,杭州,310027)       陈德钊(浙江大学,化学工程学系,浙江,杭州,310027)       胡上序(浙江大学,化学工程学系,浙江,杭州,310027)
摘    要:采用自组织映射(Self-Organize Map,SOM)网络,将复杂化学模式映射于平面,并保留模式群的高维空间拓扑结构,从而可以清晰地反映出化学模式间复杂的几何关系.还提出了获胜邻域和学习率的调整方案,用以改进SOM网络的训练效率,并将8维橄榄油样本映射于网络的输出平面,成功地实现了聚类,进而在分类预报中取得良好效果.

关 键 词:复杂化学模式  保留拓扑结构  自组织映射网络  聚类  判别  模式分类
文章编号:1008-973X(2001)05-0529-05
修稿时间:1999年10月12

Topology-preserving map of complex patterns group and its application
YAN Xue-feng,CHEN De-zhao,HU Shang-xu\,.Topology-preserving map of complex patterns group and its application[J].Journal of Zhejiang University(Engineering Science),2001,35(5):529-533.
Authors:YAN Xue-feng  CHEN De-zhao  HU Shang-xu\  
Abstract:With the development of commercializing precise analytical instrumentation and investigations of deep problems, the patterns employed to describe objectives become more complex and of higher dimension. There are many projection methods for visualizing the topology of complex high-dimensional patterns and obtaining the relationship among the patterns. Because of the outstanding good performance of the Self-Organizing Map (SOM) networks developed by Kohonen, SOM was applied to project large-dimensional complex chemical information down on two-dimensional plane, while still preserving the topology, from which the important relationships among data can be visualized easily. Algorithms for dynamically adjusting the learning rates and the size of the neighborhood around the winning unit were further proposed to improve SOM networks learning efficiency. An experiment on mapping eight-dimensional olive oil samples onto two-dimensional SOM plane and classifying them was performed successfully, and yielded satisfactory prediction.
Keywords:complex chemical information  topology-preserving  self-organizing map networks  cluster  discrimination  pattern classification
本文献已被 CNKI 维普 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号