基于双重支持向量机的网络故障诊断  

Network fault diagnosis based on Dual-SVM

作　　者：温祥西[1] 孟相如[1] 马志强[1] 

机构地区：[1]空军工程大学信息与导航学院,西安710077

出　　处：《控制与决策》2013年第4期506-510,共5页Control and Decision

基　　金：国家自然科学基金项目(61003252);全军军事学研究生课题(2011JY002-524);空军工程大学创新基金项目(201105)

摘　　要：为了提高网络故障诊断模型的建立和诊断速度,提出一种双重支持向量机(Dual-SVM)算法.通过两次SVM训练确定诊断模型:第1次SVM训练通过两类样本的类中心和样本点在类中心连线方向上的投影分布情况直接确定近似最优分类面;第2次SVM训练根据第1次的训练结果选择边界样本点并计算其模糊隶属度,通过模糊SVM获得诊断模型.在DARPA数据集上的实验表明,Dual-SVM相较SVM能够获得更快的训练速度和更加简单的故障诊断模型.The dual support vector machine（Dual-SVM） is proposed to promote the speed of establishing model and diagnosing. The diagnosis model is established by two SVM training processes. In the first process, the approximate classifying hyperplane is directly obtained by the two classes centers and the distribution of the samples on the connecting direction of centers. In the second fuzzy SVM process, the boundary samples are selected, fuzzy memberships are calculated, and the diagnosis model is established. The experiments on DARPA data-sets show that the Dual-SVM can get higher training speed and more simplified model compared to SVM.

关 键 词：网络故障诊断 支持向量机 样本选取 双重训练 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]