杭颖 廖昌义 任广为 黎楚阳

摘   要：变压器油中溶解气体方法（Dissolved Gas-in-oil Analysis， DGA）由于其结果直观、操作简单和在线观测等优势，被广泛应用于变压器领域。然而传统的DGA难以充分利用不带故障类别标签的样本信息，导致故障诊断准确率不高。针对此问题，本文提出了一种基于深度收缩自编码（Deep Contractive Auto-Encoder， DCAE）网络的变压器故障诊断方法。首先，将深度收缩自编码网络和分类器组成DCAE分类器，构建基于深度收缩自编码网络的变压器故障诊断模型。其次，考虑工程中存在的数据采样错误情况，选取部分测试样本将其随机置零，研究该模型的容错性。结果表明，所提出的变压器故障诊断模型准确率均高于传统的粒子群优化支持向量机和BP神经网络模型，该模型适用于变压器故障诊断。

关键词：变压器  油中溶解气体分析  故障诊断  深度学习  收缩自编码

中图分类号：TM41                                 文献标识码：A                        文章编号：1674-098X（2020）06（a）-0005-05

Abstract：Dissolved gas-in-oil Analysis （DGA） is widely used in the transformer field due to its advantages such as intuitive results， simple operation and online observation. However， it is difficult for traditional DGA to make full use of the sample information without fault category labels， leading to low fault diagnosis accuracy. To solve this problem， this paper proposes a transformer fault diagnosis method based on deep contractive auto-encoder （DCAE） network. First， the deep contractive auto-encoder and the classifier were combined into DCAE classifier to construct the transformer fault diagnosis model based on the deep contractive auto-encoder and dissolved gas in the transformer oil. Secondly， considering the data sampling error in engineering practice， some test samples are set to zero randomly to study the fault tolerance of the model. Finally， an example analysis shows that the accuracy of the proposed transformer fault diagnosis model is higher than that of the traditional particle swarm optimization support vector machine and BP neural network models. The model is proved to be suitable for transformer fault diagnosis.

Key Words： Power transformer; Dissolved gas-in-oil analysis; Fault diagnosis; Deep learning; Contractive auto-encoder

變压器作为电网变电过程的重要设备，其工作状态与电网安全运行密切相关[1]。变压器随着服役年限的增加，发生故障几率会显著上升，严重故障时可能导致范围性的大面积停电。目前国内存在着许多运行超过20年的油浸式变压器，及早发现这些变压器的故障，准确识别其故障类型，对于保证电网运行的可靠性具有重要意义[2]。变压器运行中因为老化、电、热故障等内部原因产生少量的气体，这些气体溶解于变压器绝缘油中，与变压器的运行状态紧密相关，可用于诊断变压器的状态。……