基于SVM集成学习的网络故障诊断方法

Network Fault Diagnosis Based on Support Vector Machine Ensemble Learning

摘要:

针对现有SVM多分类方法在网络故障诊断中识别精度较低的问题,本文提出一种基于二叉树结构和模型二重扰动的SVM集成学习算法。通过集成学习思想提高网络故障诊断的精度。在集成过程中对二叉树结构和核参数进行扰动,加大个体分类器的差异度,提升了诊断模型的泛化性。在实际网络中的诊断实验表明所提的方法较二叉树等其它SVM多分类方法具有更高的诊断精度。

Abstract:

In network fault diagnosis, the present multi-class support vector machine (SVM) classification methods have low accuracy. The ensemble learning was introduced to improve the diagnosis accuracy. A new ensemble learning algorithm based on the disturbance of both binary tree structure and model (BTSMBag) was proposed. The diversity of individual classifiers was increased by the dual disturbance and the generalization of diagnosis model was improved. The experiment on the real network fault diagnosis shows the proposed method gets higher accuracy.