基于DGA和深度置信网络的变压器内部故障诊断

Transformer internal fault diagnosis based on DGA and deep belief network

以油中特征气体组分比值为特征量的故障诊断是变压器内部故障诊断的重要方法,但实际应用中常出现"超码"和"缺码"问题,导致故障诊断精度低.从"信息驱动"的角度,提出一种基于深度置信网络的无监督型变压器故障诊断方法.该方法利用深度置信网络的油溶解气体特征提取,构建多隐含层的深度学习模型,采用无监督特征学习方法,实现在少样本情况下的变压器故障识别问题.算例表明,所提的深度置信网络提高了电力变压器故障诊断的准确率.

It is an important way to diagnose fault in power transformer using component ratio of gas in oil as a feature, but ＂over size＂ and the ＂missing code＂ problems always happen when applied, resulting in low accuracy of fault diagnosis. Considering from ＂data-driver＂, an unsupervised method of transformer fault diagnosis based on depth belief networks is proposed. Dissolved gas in oil is used as an extraction feature in deep learning network; a deep learning model including multiple hidden layers and an unsupervised learn- ing method are constructed, aiming at soving transformer failure identification problem in the case of a small sample. The example shows that the proposed method can improve the accuracy of fault diagnosis of power transformers.