极端气象条件下基于深度学习网络特征的变压器故障预测  

Transformer fault prediction based on deep learning network characteristics under extreme weather conditions

作　　者：龙玉江 姜超颖[2] 钟掖 田月炜 LONG Yujiang;JIANG Chaoying;ZHONG Ye;TIAN Yuewei(Information Center of Guizhou Power Grid Company Limited,Guiyang 550003,China;Xidian University,Xi’an 710126,China;Guiyang Power Supply Bureau,Guiyang 550001,China)

机构地区：[1]贵州电网有限责任公司信息中心,贵州贵阳550003 [2]西安电子科技大学,陕西西安710126 [3]贵阳供电局,贵州贵阳550001

出　　处：《现代电子技术》2024年第4期91-96,共6页Modern Electronics Technique

基　　金：陕西省自然科学基础研究计划项目(2022JM-336);南方电网有限责任公司科技项目(GZKJXM20200770)。

摘　　要：根据极端气象条件下变压器产生故障时的环境参数,结合变压器故障预测中常用的油中溶解气体的含量,提出一种基于深度学习网络的故障预测方法。针对已有的变压器故障诊断方法泛化能力弱、时效性低、精度低等缺点,引入极端气象参数,并通过对多组数据序列进行时因分析,提取数据随着时间的变化关系;其次,设计一种新型的神经网络,将油气参数与极端气象参数的时间特征融合,并通过深度学习网络进行故障分类与预测。仿真实验结果表明,相比于其他传统故障预测方法,所提出的极端气象条件下基于深度学习网络的变压器故障预测方法准确率有显著提高。According to the environmental parameters of transformer fault under extreme weather conditions,a fault prediction method based on deep learning network is proposed by combing the commonly used dissolved gas content in oil in transformer fault prediction.In allusion to the shortcomings of the existing transformer fault prediction methods,such as weak generalization ability,low timeliness and low accuracy,extreme meteorological parameters are introduced,and temporal analysis on multiple sets of data sequences is conducted to extract the relationship between data changes over time.A new neural network is designed to fuse the time characteristics of oil and gas parameters with extreme meteorological parameters,and fault classification and prediction are carried out by means of deep learning networks.The simulation experimental results show that in comparison with other traditional fault prediction methods,the proposed transformer fault prediction method based on deep learning network can significantly improve its accuracy under extreme weather conditions.

关 键 词：输变电变压器 故障预测 深度学习 卷积神经网络 极端气象 故障分类 溶解气体分析 

分 类 号：TN919-34[电子电信—通信与信息系统] TM407[电子电信—信息与通信工程]