基于自适应质心距投影HSVMs的网络故障识别  

Network fault identification based on ACDP-HSVMs

作　　者：张立[1] 孟相如[1] 温祥西[1] 

机构地区：[1]空军工程大学电讯工程学院,西安710077

出　　处：《计算机应用研究》2011年第11期4282-4284,4298,共4页Application Research of Computers

基　　金：陕西省自然科学基金资助项目(SJ08F14)

摘　　要：为了处理有限样本条件下的多类网络故障识别问题,提出了一种自适应质心距投影层次支持向量机。针对层次支持向量机固有的误差积累现象,该方法通过定义特征空间样本质心距投影度量类别离散程度,依据类别可分性优化偏态层次树结构;并设计基于自适应惩罚因子的补偿算法,修正由不平衡数据引起的分类超平面倾斜。实验结果表明,该方法具有较好的识别效果和效率,能够有效地抑制误差积累。This paper presented a new multi-classifier called adaptive center distance projection hierarchical support vector machine to conduct the multi-class fault pattern recognition problem with finite network sample.In order to handle the error accumulation inherent in hierarchical support vector machine,the proposed method not only optimized the structure of partial hierarchical support vector machine with class separability metic which defined by sample center distance projection in the feature space,but also corrected the skew of the optimal separating hyperplane due to the imbalanced data with compensation algorithm controlled by adaptive cost factor.Experiment results show that this method has higher identification accuracy and efficiency,and can restrain error accumulation effectively.

关 键 词：网络故障识别 支持向量机 投影距离 不平衡数据 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]