摘要： 网络故障的关联性传播可能导致网络故障数据包含大量冗余信息，影响诊断精度和处理效能。该文根据故障数据的特点，将粗糙集理论与支持向量机(SVM)相结合，采用基于邻域粗糙逼近的数值型属性约简算法进行快速高效的故障诊断，避免经典粗糙集理论中离散化误差的影响，缩减数据存储空间，降低SVM训练模型的复杂度，提高训练速度。ROC性能曲线分析结果表明，该方法具有良好的泛化能力。

关键词: 网络故障诊断, 支持向量机, 数值型属性约简, 邻域逼近

Abstract: Correlative transmitting of network faults may bring lots of redundancy information included in the network fault datas, which will affect the precision and efficiency of diagnosis. According to the characteristic of fault datas, numerical attribute reduction algorithms based on neighborhood rough approximation are adopted to carry out fast and highly efficient faults diagnosis by uniting rough set with Support Vector Machine(SVM). The discrete error in the classical RS is conquered, the memory space of data is curtailed, the complexity of SVM training model is reduced greatly, and the speed of training is put up. The well generalization of this method is analyzed and validated by ROC performance curve.

Key words: nector Machine(SVM), numerical attribute reduction, neighborhood approximation

中图分类号: 

• TP393