基于特征选择和RS-LSTM的变压器故障诊断方法

打开文本图片集

关键词：变压器；故障诊断；特征选择；长短期记忆网络；灰狼优化算法；残差连接；SE注意力机制中图分类号：TN911.23-34；TM411 文献标识码：A 文章编号：1004-373X（2025）20-0147-08

Method of transformer fault diagnosis based on feature selection and RS-LSTM

WUYuhao，ZHUWenzhong，CHENYiyue，LUO Yuan （SchoolofComputerScience&Engineering，SichuanUniversityofScience&Engineering，Yibin643O02，China）

Abstract：Inallusion totheproblemsof featureredundancyandlowclassficationaccuracyincurrent transformer fault diagnosismethodswhendealingwithmultidimensionalandnonlineardata，amethodof transformerfaultdiagnosisbasedon featureselectionandRS-LSTMisproposed.According tothelivinghabitsofwolf packs，theroledivisionofreconaissnce wolves，atack wolves，andguardwolvesareproposedtoexpandthesearchrangeofwolf packs.Thegreywolfalgorithmis improvedbycombininggatecontrol mechanismandpheromoneupdate mechanism to furtherenhance therandomnessand flexibilityoftheoptimization process.TheSEatention mechanismwithresidualconnectionsisusedtoimprove LSTM，making the modelmore focusedonkeyfeaturesrelatedtotransformerfaultsandenhancing theaccuracyoffaultdiagnosis.Bycomparing diferentmethodswithothermodels，thesuperiorityandgeneralizationof thefeatureselectionmethodwereverified.The experimentalresultsshowthat，incomparisonwithLSTM，theproposedRS-LSTMcanimproveacuracyof3.84%andcallate of 3.67% ，which provides more effctive solution for transformer fault diagnosis and has a certain engineering application value.

Keywords：transformer;faultdiagnosis;featureselection;longshort-termmemorynetwork;greywolfoptimizationalgorithm; residual connection; SE attention mechanism

0 引言

要的价值[]。（剩余8993字）