卷积神经网络在油浸式变压器故障诊断的应用  

Fault Diagnosis of Oil-immersed Transformers Based on Convolutional Neural Network

作　　者：贾茹宾 张雅君 田丰 倪艳荣 张静 JIA Rubin;ZHANG Yajun;TIAN Feng;NI Yanrong;ZHANG Jing(School of Cable Engineering,Henan Institute of Technology,Key Laboratory of Cable Structure and Materials,Xinxiang 453000,China)

机构地区：[1]河南工学院电缆工程学院,河南省线缆结构与材料重点实验室,河南新乡453000

出　　处：《电工技术》2024年第10期89-93,共5页Electric Engineering

基　　金：河南省科技攻关项目(编号232102230018,222102230062,222102240059)。

摘　　要：变压器作为变电站的主要电气设备,其智能化程度直接决定了智能变电站的发展程度,是电力系统中关系国民生产生活的重要环节。采集变压器油中溶解气体的含量及类型,通过建立卷积神经网络模型确定变压器的故障类型。在卷积神经网络算法原理的基础上,利用Java编程构建模型,将一维卷积神经网络应用到变压器故障诊断中,以变压器油中溶解的5种气体含量值作为输入向量,变压器的6种状态对应的编码值作为输出向量,并对网络中的池化层进行改进。在模型建立过程中讨论了卷积核的大小、数量、样本长度对模型精度的影响,并通过优选函数的方法确定激活函数。实验表明,将该方法生成的网络应用于变压器故障诊断,可为合理诊断变压器故障提供有价值的参考。As the main electrical equipment of substations,the smartening degree of the transformer directly determines the development degree of smart substation,and it is an important section in power system.The content and type of dissolved gas in transformer oil are collected,and the fault type of transformer is determined by establishing the convolution neural network model.On the basis of the principle of convolutional neural network algorithm,by using the java programming,one-dimensional convolutional neural network is applied to the transformer fault diagnosis through an established model.The five gas content values dissolved in transformer oil are input whereas the corresponding coding value of the six states are output.Moreover,the pooling layer in the network is improved.Then the influence of the size of the convolution kernel and the sample length on the model accuracy,and the activation function is determined by optimizing the optimal function.Experiments show that applying the network generated by this method to transformer fault diagnosis has high accuracy and can provide valuable reference for reasonable diagnosis of transformer faults.

关 键 词：变压器油气体含量 卷积神经网络 池化 故障诊断 

分 类 号：TM411[电气工程—电器]