基于张量重构融合诊断的电动汽车直流充电桩开路故障诊断方法.pdfVIP

基于张量重构融合诊断的电动汽车直流充电桩

开路故障诊断方法

ABSTRACT:Theopen-circuitfaultsofEVchargingpiles(SNR)of10dB,theaccuracyisstillabove90%,andthe

affectpowerqualityofpowergridandthreatenchargingsafety.accuracyofthismethodreaches94.54%inexperimental,

Researchesontheopen-circuitfaultdiagnosisareofgreatwhichfurtherverifiestheeffectivenessofthismethod.

significancetoensuringthesafeandstableofthepowergridKEYWORDS:DCchargingpile;faultdiagnosis;deep

andreducingthemaintenancecostofthechargingpile.learning

Accordingtothemultidimensionalcharacteristicsofopen

circuitfaultsignalsofchargingpiles,atensorreshapefusion摘要：电动汽车充电桩的开路故障影响电网电能质量、威胁

diagnosticmodelisproposedinthispaper.Inthismethod,the充电安全，研究开路故障诊断对保障电网安全稳定运行、降

multidimensionalfeatureparallelextractionabilityofresidual低充电桩维护成本具有重要意义。针对充电桩开路故障信号

network(ResNet)andthesequentialfeatureextractionability多维度特点，该文提出一种张量重构融合诊断方法。该方法

ofgatedrecurrentunit(GRU)areusedtoextractthefrontand分别利用残差网络(residualnetwork，ResNet)的多维特征并行

rearstagefaultfeaturesofthechargingpilecircuit,thefusion提取能力和门控循环单元(gatedrecurrentunit，GRU)的时序特

diagnosisisperformedforfrontandrearstagefeatures,andthe征提取能力，提取充电桩充电模块电路中前、后级故障特征，

highaccuracyfaultdiagnosisoffrontandrearstageof并对前、后级特征进行融合诊断，实现了充电桩中充电模块

chargingpileisrealized.Apre-processingmethodof前、后级故障的较高精度诊断。提出的基于张量重构的前级

front-stagethree-phasefaultsignalsbasedontensorreshape三相故障数据预处理方法，避免了传统深度学习算法使用的

methodisproposed,whichavoidsthegraphicalinputor图像化输入或一维输入，充分发挥了深度神经网络的并行诊

one-dimensionalinputusedbytraditionaldeeplearning断性能。与传统的故障诊断方法相比较，所提方法使用深度

algorithmsandgivesfullplaytothediagnosticperformanceof学习技术，无需人为选定故障特征参数。仿真证明所提方法

deepneuralnetwork.Comparedwiththetraditionalfault对不同强度噪声影响下的故障数据平均诊断准确率可达96%

diagnosismethod,theproposedmethodusesdee