基于支持向量数据描述的变压器故障检测方法

A Novel Power Transformer Fault Detection Method based on SVDD

在变压器故障诊断领域中,针对传统分类算法在没有故障样本信息的情况下无法有效实现故障检测不足的问题,提出了一种基于支持向量数据描述的变压器故障检测算法。该算法只需要目标样本,即正常样本进行训练,无需故障样本信息,最终得到描述正常样本空间结构信息的超球体,决策阶段利用与超球体中心的距离进行判断。通过实验表明,笔者提出的方法在故障样本不均衡的情况下能有效实现变压器的故障检测。

In the field of fault diagnosis of power transformer, according to the traditional classification algorithm in the absence of fault sample information can not be effective fault detection, a novel power transformer fault detection algorithm based on support vector data description. The algorithm only need to target samples, including normal sam- ples, without fault sample information, finally describes the normal sample space structure information of the hyper sphere, the decision-making stage use and hyper sphere center distance judgment. Experiments show that, the method can effectively realize the fault detection of transformer in fault samples under unbalanced condition.