摘要： 对基于径向基函数(RBF)的支持向量回归(SVR)模型参数的理论研究与实验论证结果表明，惩罚系数、不敏感损失函数的宽度以及核函数参数对非线性函数拟合精度均有影响，给出SVR参数的经验范围以减小人工选择SVR参数的盲目性，并通过缩小参数优化算法的搜索区间，降低算法的整体时间复杂度和空间复杂度。

关键词: 支持向量回归, 径向基函数, 模型参数, 非线性拟合

Abstract: The theoretical study and experimental demonstration of Support Vector Regression(SVR) based on Radial Basis Function(RBF) shows that the selection of penalty coefficient, the width of insensitive loss function and kernel function parameter can affect the accuracy of non-linear function approximation. It provides an experience range of SVR parameters to select parameters in non-linear approximation, which reduces the blindness of artificial selection parameters and narrows the search range of parameters optimization algorithm to reduce algorithm whole time complexity and space complexity.

Key words: Support Vector Regression(SVR), Radial Basis Function(RBF), model parameter, non-linear approximation

中图分类号: 

• TP181

• TP301