基于最小二乘模糊单类支持向量机的网络故障检测

Network fault detection based on fuzzy one class SVM with least squares and equality constraints

针对基于单类支持向量机的网络故障异常检测存在的训练速度慢和检测精度低等问题,提出一种最小二乘模糊单类支持向量机(LSFOC-SVM)。该方法采用最小二乘损失函数和等式化约束改进标准单类支持向量机的训练算法,将二次规划转化为解线性方程组,降低了计算代价;并通过构造基于特征空间距离的模糊隶属度函数和优化选择告警阈值,适当扩大了故障预警范围,提高了故障检测率。与同类方法相比,该方法在保证检测效果的同时大幅度地提升了训练效率。应用测试结果表明该方法是可行的。

A new classifier named Least Squares Fuzzy One Class Support Vector Machine （LSFOC-SVM） was proposed to enhance the efficiency and effect of one class support vector machine applied to network fault abnormal detection. The proposed LSFOC-SVM not only reduced the high computational cost by training with the least squares and equality constraint which obtain a set of linear equations instead of quadratic programming, but also enhanced the fault detection rate by extending the fault alarm area properly with fuzzy membership based on distance in feature space and appropriate alarm threshold. The comparative study results indicate LSFOC-SVM can improve the training efficiency greatly without affecting the diagnosis accuracy. And application tests verify the feasibility of this method.