边界偏转覆盖增量支持向量机  

Boundary Deflection Overlay Incremental Support Vector Machine

作　　者：张立[1] 孟相如[1] 马志强[1] 周华[2] 

机构地区：[1]空军工程大学电讯工程学院,西安710077 [2]西安通信学院,西安710106

出　　处：《北京邮电大学学报》2010年第4期30-34,共5页Journal of Beijing University of Posts and Telecommunications

基　　金：陕西省自然科学基金项目(SJ08F14)

摘　　要：为了利用不断积累的网络样本提高故障诊断效能,针对标准支持向量机不直接支持增量学习的问题,提出一种边界偏转覆盖增量支持向量机.根据违背Karush-Kuhn-Tucker条件的新增样本在特征空间中可引起原分类边界改变的情况,设计边界偏转覆盖算法预选支持向量再生区作为增量训练工作集,解决了难以确定的非支持向量向支持向量的转化问题.理论分析和实验结果表明,该方法能有效简化训练工作集,在保证故障诊断精度的同时大幅度提高增量训练效率.In order to enhance the diagnosis efficiency with the accumulated network sample,and because the standard support vector machine doesn＇t support incremental learning directly,a boundary deflection overlay incremental support vector machine is proposed. According to the movement of separating hyperplane caused by the newly added training samples that violate Karush-Kuhn-Tucker conditions,the boundary deflection overlay algorithm is also designed to pre-extracts support vector reproducing region as the work set for incremental training,which solves the problem that non-support vectors transform to new support vectors. Analysis and simulation show that the method can not only reduce the work set effectively,but improve the training efficiency greatly without affecting the diagnosis accuracy.

关 键 词：故障诊断 支持向量机 增量学习 模型更新 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]