两种基于等式约束支持向量回归机比较研究  

Comparison of Two Types of Support Vector Regressions Based on Equality Constraint

作　　者：蒋蔚[1] 伊国兴[1] 曾庆双[1] 

机构地区：[1]哈尔滨工业大学航天学院,黑龙江哈尔滨150001

出　　处：《控制工程》2009年第6期743-747,共5页Control Engineering of China

基　　金："十一五"国防预研基金资助项目(51309030102);"十一五"国防预研基金资助项目(51309030203)

摘　　要：为了减小支持向量回归机(SVR)的计算复杂度、缩短训练时间,将应用于分类问题的近似支持向量机(PSVM)扩展到回归问题中,针对其原始优化问题采用直接法求取最优解,而不是转换为对偶问题求解,给出了近似支持向量回归机(PSVR)线性和非线性回归算法。并与同样基于等式约束的最小二乘支持向量回归机(LSSVR)进行了比较,在一维、二维函数回归以及不同规模通用数据集上的测试结果表明,PSVR算法简单,训练速度快,尤其在大规模数据集处理上更具优势。To reduce computation complexity and training time of the support vector regression, the proximal support vector machines used in classification problems is extended into regression problems. And the optimal solution is obtained directly from the primitive problem rather than the dual problem. The linear and nonlinear proximal support vector regression algorithms are presented and compared with the least squares support vector regression based on equality constraint. The numerical results based on l D and 2D regression and different-scale public data sets demonstrate that the proximal support vector regression is a simple and fast training algorithm especially for large scale data sets.

关 键 词：近似支持向量回归机 最小二乘 等式约束 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]