一种新的多类支持向量机决策方法及其应用  

A Novel Decision-making Method for Multi-class SVMs and Its Application

作　　者：王晓红[1] 

机构地区：[1]九江学院江西省数控技术与应用重点实验室,江西九江332005

出　　处：《信息与控制》2008年第6期647-652,共6页Information and Control

基　　金：国家自然科学基金资助项目(50705039)

摘　　要：讨论和比较了现有的几种多类SVM方法.在此基础上,提出了一种组合多个两类分类器结果的多类SVM决策方法.在该方法中,定义了新的决策函数,其值是在传统投票决策值的基础上乘以不同分类器的权重.新的多类SVM在一定程度上解决了传统投票决策方法的不可分区域问题,因此具有更好的分类性能.最后,将新方法作为关键技术应用于故障诊断实例,实际诊断结果证明了所提多类SVM决策方法的优越性.This paper presents several existing multi-class classifiers and compares their advantages and disadvantages. Then, a decision-making method for multi-class SVMs is proposed to recombine the outputs of several binary classifiers. In the strategy, a novel decision function is defined, and its value is the traditional voting values multiplied by the corresponding weights of different classifiers. The presented multi-class SVM is of better classification ability and can solve the unclassifi- able region problems in traditional max-wins-voting （MWV） strategy. Lastly, the novel method is used in fault diagnosis as the key technology, and the actual diagnosing results demonstrate the superiority of the improved strategy over the traditional one.

关 键 词：多类支持向量机 投票法 策略 不可分区域 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]