论文部分内容阅读
支持向量机通过引入核函数将低维空间的非线性问题转化为高维空间的线性问题,克服了维数灾难,并展现了极好的学习能力。但是在支持向量回归分析中,核函数的选取和模型参数的选择目前都没有十分有效的方法。针对高斯核函数的情况,首先通过理论分析和数值仿真,给出了模型参数的选取范围,然后结合均匀试验设计和偏最小二乘回归,提出了一种快速有效的模型参数选择方法。理论分析和实例计算表明该方法选取的模型参数确实能够得到泛化能力较好的回归模型。
Support vector machine (SVM) overcomes the dimensionality disaster by introducing the kernel function and transforms the nonlinear problem of low-dimensional space into the linear problem of high-dimensional space, and shows excellent learning ability. However, in the support vector regression analysis, the choice of kernel function and the choice of model parameters are not very effective at present. In the case of Gaussian kernel function, the selection range of the model parameters is given firstly through theoretical analysis and numerical simulation. Then, a fast and effective model parameter selection method is proposed based on uniform experimental design and partial least-squares regression. Theoretical analysis and case calculation show that the model parameters selected by this method can indeed get the regression model with good generalization ability.