支持向量机最优参数选择的研究  

Study on the choice optimum parameters of support vector machine

作　　者：刘东辉[1] 卞建鹏[1] 付平[1] 刘智青[1] 

机构地区：[1]河北科技大学电气信息学院,河北石家庄050018

出　　处：《河北科技大学学报》2009年第1期58-61,共4页Journal of Hebei University of Science and Technology

基　　金：河北省自然科学基金资助项目(F2007000636)

摘　　要：支持向量机是近几年发展起来的机器学习方法,模型选择是设计支持向量机的关键。基于高斯核函数的支持向量机具有良好的学习性能,被广泛应用于模式分类中,讨论了核函数中C和γ对分类函数的影响,提出针对不同类型的数据,SVM应选用不同的核函数,同时利用二分法对核函数(C,γ)寻优,并将其应用于变压器故障诊断中,仿真结果表明该方法具有较好的性能。Support vector machine （SVM） is a novel machine learning method. Model selection is essential to design SVM. The support vector machine based on Gauss kernel function is widely used in model classification due to its good properties. This paper studied the influences of the error penalty parameter and the kernel parameter on support vector machine＇s generali- zation ability and the method of the choice of the optimum parameter at present. At the same time, this paper presented that SVM should choose different kernel functions in terms of the different dataset, and the two-divided method is proposed for the choice of the Optimum parameter. The method presented by the paper was applied to the fault diagnosis of the oil-immersed transformer. The simulation experiment show that this method has good performance.

关 键 词：支持向量机 模型选择 高斯核函数 二分法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]