文档摘要：冲击电压老炼技术是提高真空电极间隙绝缘能力的有效手段,快速准确地甄别真空击穿机制对揭示冲击电压老炼过程的物理演化机理具有重要意义.该文提出了一种基于深度学习的通过突显击穿前过程提高真空击穿机制甄别准确度的优化方法.对五对相同的无氧铜球形电极开展同样的冲击电压老炼试验,分别获得时间提取范围为0-400μs的完整击穿电压电流波形样本和0-200μs的突显击穿前过程的击穿电压电流波形样本,并通过深度学习模型对两种击穿电压电流波形样本开展脉冲电流诱发、场致发射诱发和微粒诱发三种真空击穿机制的甄别训练与测试,并将测试结果与真实结果进行对比分析与评估.结果显示:时间提取范围为0～200μs的突显击穿前过程的击穿电压电流波形样本的击穿机制甄别准确率均在87.99％以上,平均提高了2.55％,其对应的精确率、召回率和F1分数均更优.研究结果表明,突显击穿前过程的击穿电压电流波形处理能够有效地优化真空击穿机制甄别的效果,具有良好的工程应用前景.
Abstract：Impulsevoltageconditioningtechnologyisaneffectivemeanstoimprovetheinsulationabilityofvacuumcircuitbreaker(VCB).ClassifyingthebreakdownmechanismquicklyandaccuratelyhasagreatsignificancetorevealthephysicalevolutionofimpulsevoltageconditioningandimprovetheVCBwithstandingvoltagelevel.ThetraditionalmethodtoclassifythebreakdownmechanismneedstoeliminatethedisplacementcurrentthroughmathematicalcompensationalgorithmandfittheFowler-Nordheimformula,whichiscomplicatedtoobtainthebreakdownmechanism.Deeplearninghasanobviousadvantageinimagerecognitionandfeatureextraction.Inthispaper,anoptimizedmethodtoclassifythebreakdownmechanismwasproposedthroughenlargingthepre-breakdownperiodinbreakdownwaveformbasedondeeplearning.Fiveidenticalsphereoxygen-freecopperelectrodepairsA,B,C,DandEwereappliedthesameimpulseconditioning.Allthebreakdownwaveformswereprocessedintotwokinds:0～400μs,containingthewholebreakdownwaveform,and0～200μs,pre-breakdownperiodenlargedbreakdownwaveform.ThecorrespondingbreakdownmechanismsofAandBwerelabeledaspulsedcurrentinducedvacuumbreakdown(PB),fieldemissioninducedbreakdown(FEBD)andparticleinducedvacuumbreakdown(PBD)throughthetraditionalmethod.Then,breakdownwaveformsofAandB(1530)in0～400μsand0～200μswereforthebreakdownmechanismclassificationtraining,andbreakdownwaveformsofC,DandE(1398)in0～400μsand0～200μswereforbreakdownmechanismclassificationtest,respectively.ThecorrespondingbreakdownmechanismsofC,DandEwereclassifiedintoPB,FEBDandPBDwithdeeplearning.Inaddition,thebreakdownmechanismsofC,DandEwerealsoobtainedthroughthetraditionalmethod.Thedeeplearningoutputswerecomparedwiththatthroughthetraditionalmethod.Thetestresultswereevaluatedandanalyzedbytheevaluationparameterssuchasprecision,recall,Fl-scoreandsoon.TheresultsshowedthatthebreakdownmechanismclassificationaccuraciesofC,DandE(0～200μs)were88.92％,87.99％and92.78％,respectively,andalltheaccuraciesof0～200μswerehigherthan87.99％.ThebreakdownmechanismclassificationaccuraciesofC,DandE(0～400μs)were85.23％,84.90％and91.90％,respectively.Comparedwith0～400μs,thebreakdownmechanismclassificationaccuraciesof0～200μswereimprovedby3.69％,3.09％and0.88％,respectively.Theaccuracyof0～200μshadanaverageimprovementby2.55％thanthatof0～400μs.Precision,recallandFl-scoreof0～200μswerealsohigherthanthoseof0～400μs.Theresultsshowedthat0～200μs,pre-breakdownperiodenlargedbreakdownwaveformhadabetterperformanceinbreakdownmechanismclassification.Conclusionsweredrawnasfollowing:(1)Theclassificationaccuracyforbreakdownmechanismthroughdeeplearningcouldbeimprovedbyenlargingthepre-breakdownperiodinthebreakdownwaveform.(2)Thebreakdownmechanismclassificationcanbecompletedquicklyandaccurately,whoseaccuracycouldbehigherthan87.99％withtheeffectivenessverifiedbyprecision,recallandFl-score.IthasatheoreticalguidanceforapromisingconditioningtechnologytoimprovetheVCBvoltagelevelinindustryapplication.
作者：李世民 徐勋晨 张潮海Author：LiShimin XuXunchen ZhangChaohai
作者单位：南京航空航天大学多电飞机电气系统工信部重点实验室南京210016
刊名：电工技术学报 ISTICEIPKU
Journal：TransactionsofChinaElectrotechnicalSociety
年，卷(期)：2024, 39(13)
分类号：TM561.2
关键词：冲击电压老炼 击穿机制 深度学习 突显击穿前过程 击穿波形
Keywords：Impulsevoltageconditioning breakdownmechanism deeplearning pre-breakdownperiodenlarged breakdownwaveform
机标分类号：TM835O634.41TM131.41
在线出版日期：2024年7月22日
基金项目：国家自然科学基金，江苏省自然科学基金，中央高校基本科研业务费专项资金资助项目，中央高校基本科研业务费专项资金资助项目基于深度学习的冲击电压老炼过程中真空击穿机制甄别优化方法[
期刊论文] 电工技术学报--2024, 39(13)李世民 徐勋晨 张潮海冲击电压老炼技术是提高真空电极间隙绝缘能力的有效手段,快速准确地甄别真空击穿机制对揭示冲击电压老炼过程的物理演化机理具有重要意义.该文提出了一种基于深度学习的通过突显击穿前过程提高真空击穿机制甄别准确度的...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
关键词：冲击电压老炼,击穿机制,深度学习,突显击穿前过程,击穿波形,
- 文件大小:
- 13.02 MB
- 下载次数:
- 60
-
高速下载
|
|
1
