摘要： 传统的支持向量机是基于两类问题提出的，如何将其有效地推广至多类分类仍是一个研究的热点问题。在分析比较现有支持向量机多类分类OVO方法存在的问题及缺点的基础上，该文提出一种新的基于概率投票策略的多类分类方法。在该策略中，充分考虑了OVO方法中各个两类支持向量机分类器的差异，并将该差异反映到投票分值上。所提多类支持向量机方法不仅具有较好的分类性能，而且有效解决了传统投票策略中存在的拒分区域问题。将基于概率投票的多分类支持向量机作为关键技术应用于实际齿轮箱故障诊断，并与传统投票策略的结果进行对比，表明所提方法的上述优点。

关键词: 支持向量机, 多类分类, 概率, 故障诊断

Abstract: Traditional Support Vector Machine(SVM) is originally designed for binary classification. How to effectively extend it to multi-class classification is still an on-going research issue. After analysis and comparison of the problems and defections of the existing One-Versus-One(OVO) methods of multi-class SVMs, the novel multi-class classification method based on probability voting strategy is put forward. In the new strategy, the differences and different weights among these two-class SVM classifiers are considered and combined with the value of voting. The presented multi-class SVMs method can achieve the better classification ability and resolve the unclassifiable region problems in the conventional Max-Wins-Voting(MWV) strategy. Moreover, the multi-class SVMs method based on probability voting is used as the key technology of fault diagnosis for gearbox. The practical results show that compared with traditional MWV strategy, the presented one is effective.

Key words: Support Vector Machine(SVM), multi-class classification, probability, fault diagnosis

中图分类号: 

• TP182