Abstract: A new method based on Support Vector Machines with the principle of Structural Risk Minimization are proposed.Theoretical analysis proved,this method can successfully solve the problems in the application of BP neural network including overfit and the danger of getting stuck into local minima,also it avoids the information loss occurred in the application of Primary Component Analysis.Experimental results show that compared to BP Neural Network,the training process of Support Vector Machines is more controllable;and the relative error of evaluation score based on support vector regression machines is smaller;Furthermore,the evaluation differences of the samples are maintained better.

Key words: Support Vector Machine（SVM）, BP Neural Network, Primary Component Analysis（PCA）, empirical risk minimization, structural risk minimization

摘要： 为了实现移动网络各粒度网元的自动综合评价,针对现有的移动网络性能综合评价方法在应用中存在的问题,提出了一种新的以结构风险最小原则为理论基础的支持向量机评价方法。理论分析表明,该方法可以克服BP神经网络的评价方法中存在的过拟合以及可能收敛于局部极小点的问题;也避免了主成份分析法导致的信息丢失问题。实验结果表明,采用基于支持向量机的评价方法比之基于BP的方法,预测误差更小,过程更可控,而且更好地保持了不同样本评价间的差异。

关键词: 支持向量机, BP神经网络, 主成份分析, 经验风险最小, 结构风险最小