虚实特征融合与数据-机理驱动的变压器绕组小样本故障诊断

Small sample fault diagnosis of transformer windings based on virtual-real feature fusion and data-mechanism driven

针对变压器绕组诊断的小样本场景,提出一种虚实特征融合的数据-机理驱动方法,通过引入仿真模型弥补实际样本特征不足的问题。提出虚实特征融合方法的原理,推导其基本公式,构建虚实特征融合的小样本故障诊断框架。搭建变压器绕组故障测试实验平台,采集包含不同故障位置、故障类型、故障程度的测试实验样本集,同时获得仿真模拟虚拟数据集。对比试验表明,所提数据-机理融合方法对多种智能算法在不同小样本程度下的诊断性能均有提升效果:在小样本程度为30%的小样本场景下,基于所提方法的平均诊断准确率提高了27.1%。针对虚实数据相似度对诊断结果的影响进行了探讨。

Aiming at the small sample scenario of power transformer winding fault diagnosis,a data-mechanism driven method of virtual-real feature fusion is proposed,which introduces the simulation model to make up for the lack of actual sample features.The principle of virtual-real feature fusion method is proposed,its basic calculation formula is deduced,and a small sample fault diagnosis framework of virtual-real feature fusion is constructed.The transformer winding fault test experimental platform is established,and test sample sets containing samples of different fault locations,fault types and fault degrees are collected.Meanwhile,the corresponding simulation data set is obtained.Through comparative experiments,it is verified that the proposed data-mechanism driven method can improve the diagnostic performance of various intelligent algorithms at different small sample levels.Under the small sample scenario with a small sample level of 30%,the average diagnostic accuracy based on the proposed method is improved by 27.1%.The influence of the similarity of virtual and real data on diagnosis results is discussed.