基于模糊机会约束的超球支持向量机  

An hyper-sphere SVM based on fuzzy chance constraint

作　　者：周绍磊[1] 秦亮[2] 史贤俊[1] 肖支才[1] 

机构地区：[1]海军航空工程学院控制工程系,山东烟台264001 [2]海军航空工程学院研究生管理大队,山东烟台264001

出　　处：《华中科技大学学报（自然科学版）》2012年第7期29-33,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61004002)

摘　　要：针对不确定数据多分类问题,提出一种基于模糊机会约束的超球支持向量机(FCC-HSVM).在球结构支持向量机的基础上,引入模糊事件的可能性测度,得到模糊机会约束规划及其对偶规划.利用球结构的优点,每类样本只参与一个超球体的训练,直接求解多分类模型,提出FCC-HSVM的快速学习算法,显著缩短多分类情况下训练时间.数据试验表明:这种支持向量机分类精度较高,训练速度快,适合解决不确定数据多分类问题.Aiming at the problem of multi-class pattern recognition for uncertain data,a hyper-sphere structure support vector machine（FCC-HSVM）based on fuzzy chance constraint was proposed.The possibility measure was introduced into hyper-sphere structure SVM and the problem was formulated as a fuzzy chance constrained programming.This classifier was used to deal with uncertain data and its training speed was higher as each category data trained only one sphere.Thus,a fast training method based on SMO（sequential minimal optimization）was developed to obtain the result.Numeric experiments show that the accuracy and speed of classification can be improved which is suitable for practical use.

关 键 词：机器学习 模糊机会约束 超球支持向量机 多分类 快速学习算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]