本文对支持向量机用于回归估计进行了详细的性能分析,得出了不敏感系数、惩罚因子和核函数及其参数是影响支持向量机回归估计性能的主要因素。不敏感系数可控制模型的泛化推广能力,其值的确定应考虑样本可能带有的噪声分布状况,惩罚因子可控制拟合曲线复杂性,核函数宽度系数可影响回归曲线光滑程度。因此,在采用支持向量机回归建模时,应根据建模对象选定合适的参数值,以保证回归建模效果。最后通过对原油实沸点蒸馏曲线的拟合问题验证了分析结果,为进一步研究确定SVM参数的优化方法打下了基础。 In this paper, the performance of support vector machine for regression estimation is analyzed in detail, and the insensitivity coefficient, penalty factor, kernel function and its parameters are the main factors that affect the performance of SVM regression estimation. The insensitive coefficient can control the generalization ability of the model. The value should be determined by considering the possible noise distribution in the sample. The penalty factor can control the complexity of the fitted curve. The kernel function width factor can affect the smoothness of the regression curve. Therefore, when using support vector machine regression modeling, the appropriate parameter values ​​should be selected according to the modeling object to ensure the regression modeling effect. Finally, the results of the analysis are validated by the fitting of the real boiling point distillation curve, which lays the foundation for further research on the optimization of SVM parameters.