支持向量回归估计性能分析  

Regression performance analysis of support vector regression

作　　者：宋晓峰[1] 陈德钊[2] 胡上序[2] 

机构地区：[1]南京航空航天大学生物医学工程系,江苏南京210016 [2]浙江大学智能信息工程研究所,浙江杭州310027

出　　处：《计算机与应用化学》2005年第7期495-499,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(20076041)

摘　　要：本文对支持向量机用于回归估计进行了详细的性能分析,得出了不敏感系数、惩罚因子和核函数及其参数是影响支持向量机回归估计性能的主要因素。不敏感系数可控制模型的泛化推广能力,其值的确定应考虑样本可能带有的噪声分布状况,惩罚因子可控制拟合曲线复杂性,核函数宽度系数可影响回归曲线光滑程度。因此,在采用支持向量机回归建模时,应根据建模对象选定合适的参数值,以保证回归建模效果。最后通过对原油实沸点蒸馏曲线的拟合问题验证了分析结果,为进一步研究确定SVM参数的优化方法打下了基础。The performances of support vector regression were analyzed. It was found that the insensitive factor, penalty factor C and the kernel function along with its parameters are the main factors affecting the performance of support vector regression estimation. The insensitive factor can control the support vector regression generalization performance, and should be determined according to the noise in sample data. Penalty factor can control the curve complexity. The width factor of kernel function can affect the curve smoothness. Therefore, to obtain the good performance of support vector regression estimation, the parameters of support vector regression should be properly selected firstly when support vector machine are employed for modeling. All of these works can establish the basis of parameters optimization in support vector regression. In the end, test example was employed to validate our analyzing results.

关 键 词：支持向量回归 回归估计 参数 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]