基于FedAVG-SA-BP的变压器DGA故障诊断  

Transformer DGA Fault Diagnosis Based on FedAVG-SA-BP

作　　者：周威振 马越 邓集瀚 毛玉星[2] 黄澳丽 ZHOU Wei-zhen;MA Yue;DENG Ji-han;MAO Yu-xing;HUANG Ao-li(Dali Bureau,China Southern Power Grid Co.,Ltd.,Dali 671000,China;School of Electrical Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]南方电网超高压输电公司大理局,云南大理671000 [2]重庆大学电气工程学院,重庆400044

出　　处：《变压器》2023年第9期31-35,共5页Transformer

摘　　要：本文中作者提出了基于FedAVG-SA-BP方法的变压器油中气体分析(Dissolved Gas Analysis, DGA)故障诊断模型,采用Self-attention(SA)机制捕捉气体特征间的相关性,以及Federated Averaging(FedAVG)解决变压器DGA故障诊断中训练样本缺失的问题。试验结果表明,与传统的BP网络DGA故障诊断方法相比,该方法能够显著提高电力变压器故障诊断的准确性。This paper proposes a transformer oil dissolved gas analysis(DGA)fault diag-nosis model based on FedAVG-SA-BP,which adapts the self-attention(SA)mechanism to capture the correlation between gas characteristic,and the federated averaging(Fe-dAVG)to deal with the insufficient training sample issue in DGA fault diagnosis.The ex-perimental results suggest that the proposed method can significantly improve the DGA diagnosis accuracy comparing with the traditional BP-based methods.

关 键 词：变压器 油中气体分析法 BP神经网络 

分 类 号：TM411[电气工程—电器]