【摘要】 尽管支持向量机在许多问题上有着良好的表现,但是其参数和核函数的参数选取问题依然亟待解决。以往多采用优化算法进行参数选取,但也需要预先经验地获得核函数的参数的选取范围。在介绍结构风险最小化原则及支持向量机算法的基础上,给出了基于优化算法的支持向量机参数选取的一般性算法。由于径向基核函数(RBF)的参数取值大小的不同,可导致其性质和作用不同,为此提出了一种分段函数对RBF的参数进行选择的方法,该方法使得RBF的参数取大值和小值的概率均等。由此可不必预先经验地指定RBF的参数的选取范围,依然可以优化获得最优的参数。通过对头部组织电导率估算问题进行对比研究,取得了良好的效果,验证了该方法的有效性。更多还原

【Abstract】 Although Support Vector Machine(SVM) performs well in many situations,it is still difficult to select its parameter and the parameter of kernel function.Previously,optimisation algorithms are mostly used for finding better parameters of SVM and its kernel function,however the selection range of the parameters of kernel function also need to empirically obtain.The paper introduces the Structural Risk Minimisation(SRM) principle and SVM algorithm.Based on that,a optimisation algorithm(OA) based general algorithm for SVM parameters selection is brought forward.Since different value-taking in parameters of Radial Based Function(RBF) can cause difference in effects and properties,so we propose a piecewise function method for selecting the parameter value of RBF,which makes equal probability in selecting bigger or smaller RBF parameter value.Hence the selection range of RBF parameters could not be empirically appointed but the optimum parameter can be still got in an optimised way.At last,the comparative study was carried out on head tissue conductivity estimation and achieved a good outcome,which proved the effectiveness of the method.更多还原