【作者】 李新燕；

【导师】 汪梅；

【作者基本信息】 西安科技大学 ， 模式识别与智能系统， 2010， 硕士

【摘要】 支持向量机是一种专门针对小样本的模式识别方法,它建立在统计学习理论的VC维和结构风险最小化原理的基础上,较好地解决了非线性、高维数、局部极小点等问题,具有良好的推广性。经典的支持向量机是一种两类分类器,随着应用范围的扩展,支持向量机被广泛地用于解决多类分类问题。如何针对具体的问题建立较好的支持向量机多分类模型,是近几年支持向量机研究的热点之一。本文在统计学习理论相关原理及支持向量机理论基础上,对支持向量机多类分类算法及核函数的参数选择问题进行了分析与研究。论文概述了统计学习理论、支持向量机的分类原理、支持向量机集成的最优分类面、分类间隔、最优化理论等技术,介绍了支持向量机的重要模块——核函数,并对不同核函数的性能进行了比较;分析和总结了现有的几种支持向量机多类分类方法,包括一次性求解算法、“一对多”、“一对一”、有向无环图算法等,比较了它们的性能及优缺点;针对现有的支持向量机多类分类算法中存在错分或拒分区域问题和二叉树支持向量机结构的生成问题,在定义了较小邻近距离和构造类间相异度矩阵的基础上,提出了基于类间相异度矩阵的支持向量机多类分类算法,并通过仿真实验,与传统的一对一、一对多及有向无环图支持向量机算法结果进行比较,验证了该算法的正确性和有效性。论文最后部分将基于类间相异度矩阵的支持向量机多分类算法用于对电缆故障类型的识别分析中,通过尝试不同的核参数选择出较优的核参数以建立支持向量机多类识别模型,实现了对故障类型有效的识别。更多还原

【Abstract】 Support vector machine is a pattern recognition method specially for small samples, which is based on the VC dimensions of the statistical learning theory and the structural risk minimization principle. It well sloves the problems of nonlinear, high dimensions and local minimum. It has a good generalization performance. The classical support vector machine is a two classs classifiers. With the expansion of its application,it is widely used to solve the multi-class classification problems. For a special problem, how to build a better support vector machine multi-classification model has become one of the spots of support vector machines in recent years.The multi-class support vector machine algorithm and the selecting problem of kernel parameters are analyzed and researched in the thesis based on the statistical learning theory and the support vector machine theory. First, the statistical learning theory, the support vector machine classification principle and the techniques which the support vector machine has integrates including the optimal hyperplane, the classification interval, the optimization theory are summarized. At the same time, the kernel function, which is an important module of the support vector machine is described and the performance of differnt kernel functions are compared. Then, several existing multi-class support vector algorithms including the one-time algorithm, one versus rest algorithm, one versus one algorithm and the acyclic graph algorithm are analyzed and their performance, advantages and disadvantages are compared. Aming at the problem that there exsits some wrongly classified region or refused region in the above algorithms and the difficulty to determine the structure of the binary tree in the binary tree support vector machine algorithm, a support vector machine algorithm based on the between-class dissimilarity matrix is presented after the two novel concepts of the shorter neighboring distance and the between-class dissimilarity matrix are defined. In the simulation, the results of the new algorithm are compared with that of one versus one algorithm, one versus rest algorithm and the acyclic graph algorithm, which verifies that the alogrithm is correct and effective.In the last part of the thesis, the support vector machine algorithm based on the between-class dissimilairity matrix is applied to the recognition for the types of cable fault. The fault types are finally effectively recognized after selecting a better parameter by trying differernt kernel parameters to build the multi-class support vector machine model.更多还原