基于粗糙集与多类支持向量机的电力变压器故障诊断

Fault Diagnosis for Power Transformer Based on Rough Set and Multi-class Support Vector Machine

针对传统变压器故障诊断过程中未能将部分反映变压器故障状态的信息有效利用,以致故障诊断信息不完备、诊断结果不准确的情况,将铁芯接地电流等信息与特征气体相结合,以完善故障特征信息。并在此基础上,构建了一种采用粗糙集的一对一多类支持向量机故障诊断新方法。首先利用一对一多类支持向量机实现故障类别区域的划分;然后根据粗糙集的上下近似这一核心思想对故障类别划分区域进行描述,得出故障分类的上下近似域及边界域的集合,并提取故障诊断分类规则;最后利用分类规则实现故障类别划分。该方法实现了故障信息的综合利用,并将粗糙集在不完备数据与复杂模式刻画方面所具备的优良表现,及一对一支持向量机在分类方面的良好泛化性能进行有效融合,从而有效提高故障分类精度。变压器故障实例分析表明,与传统诊断方法相比较,该方法具有更高的诊断正确率,且其可有效反映故障诊断中所出现的不完备信息。

Some of the characteristic parameters are not effectively used in conventional fault diagnosis of transformer, which will cause the inaccuracy of diagnostic results. To improve the present situation, the core grounding current infor- mation is proposed to be combined with characteristic gases dissolved in oil as the input variables of one-against-one multi-class support vector machine in transformer fault diagnosis. Firstly, one-against-one multi-classes support vector machine is used to divide the fault category region. Secondly, the fault category region is described according to upper and lower approximation of rough set theory. Thirdly, the upper and lower approximation domain and the boundary do- main of the fault classification are obtained, and the fault diagnosis classification rules are extracted. Finally, the attracted classification rules are used to realize fault classification. The proposed method realized integrating fault information, comprehensively utilizing the advantage of the performance of rough set theory in processing incomplete data, complex pattern depiction, and the advantage of good generalization performance of one-to-one support vector machine in the classification, so as to effectively improve the accuracy of fault classification. An example of transformer failure analysis show that the proposed method has higher diagnostic accuracy, and it can effectively reflect the incomplete information in the fault diagnosis.