支持向量机最优参数选择的研究-河北科技大学学报.PDF

30 1 Vol. 30, No . 1 2009 3 Journal of Hebei University of Science and T echnolo y Mar . 2009 : 2009) 刘东辉, 卞建鹏, 付 平, 刘智青 ( 河北科技大学电气信息学院, 河北石家庄 050018) : 支持向量机是近几年发展起来的机器学习方 , 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 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 对核函 数( C, ) 寻优, 并将其应用于变压器故障诊断中, 仿真结果表明该方 具有较好的性能 : 支持向量机; 模型选择; 高斯核函数; 二分 : T P18 : A Study on the choice optimum parameters of support vector machine LIU Don hui, BIAN Jianpen , FU Pin , LIU Zhiqin ( Colle e of Electrical En ineerin and Information Science, Hebei University of Science and T echnolo y, Shijiazhuan H ebei 050018, China) Abstract: Support vect or machine ( SV M) is a novel machine learnin method. Model selection is essent ial to desi n SV M . The support vector machine based on Gauss kernel function is w idely used in model classification due t o its ood properties. This paper studied t he influences of the error penalt y parameter and the kernel parameter on support vector machines enerali zation ability and the met hod of the choice of the opt imum 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 twodivided method is proposed for the choice of the optimum parameter. T he method present ed by the paper w as applied to the fault dia nosis of the oilimmersed transformer . T he simulation experiment show that this method has ood performance. Key words: support vector machine; model choice; Gauss kernel function; twodivided method ( support vector machine) 20 90 V apnik , , : 1) , , ; 2) , , , , , , , : : : : ( F2007