一维全卷积神经网络在变压器故障诊断中的应用研究  

Application of one-dimensional all convolutional neural network in fault diagnosis of power transformer

作　　者：谭佳文 张文韬 王雅琪 TAN Jiawen;ZHANG Wentao;WANG Yaqi(China Yangtze Power Co.,Ltd.)

机构地区：[1]中国长江电力股份有限公司三峡水力发电厂,湖北宜昌443133

出　　处：《大坝与安全》2025年第1期61-66,共6页Dam & Safety

摘　　要：为提高变压器故障诊断的准确性,提出了一种基于一维全卷积神经网络的油浸变压器故障诊断模型。首先对变压器油中溶解气体数据进行标准化处理和编码;然后构建上述诊断模型的网络,网络结构运用全卷积网络和卷积核的特点,能够有效提取输入数据中的故障特征并用于分类;最后对模型进行了训练和测试。试验结果及对比分析表明,上述方法应用于电力变压器故障诊断时具有较高的准确性。To improve the accuracy of fault diagnosis for transformer,a fault diagnosis model for oil immersed transformer based on one-dimensional all convolution neural network is proposed.Firstly,the data of gases dissolved in the transformer oil is standardized and coded.Then the network of the proposed diagnosis model is established.By using the characteristics of all convolutional network and convolution kernel,the network structure can effectively extract fault features from input data and then apply it for classification.Finally,the model is trained and tested.The test result and contrastive analysis show that superior performance is achieved when the model is applied to fault diagnosis of power transformer.

关 键 词：卷积神经网络 深度学习 变压器 故障诊断 溶解气体分析 

分 类 号：TM411[电气工程—电器]