电力系统及其自动化学报ProceedingsoftheCSU-EPSA第35卷第2期2023年2月Vol.35No.2Feb.2023光伏并网逆变器参数性故障的VMD-WPE和MPA-LSTM诊断方法研究张彼德，余海宁，罗荣秋，张锦，冯京（西华大学电气与电子信息学院，成都610039）摘要：针对三相光伏并网逆变器参数性故障的特征量提取难、诊断准确率较低等问题，提出变分模态分解的小波包能量特征与海洋捕食者算法优化长短期记忆神经网络相结合的故障诊断方法。首先，以逆变器三相线电压为原始数据，以最小样本熵为准则优化变分模态分解的模态数；然后，利用小波包分解提取变分模态分解各模态分量的小波包能量作为故障特征量；最后，利用海洋捕食者算法优化长短期记忆网络超参数实现故障的参数性辨识。对比分析结果表明，所提方法用于光伏并网逆变器参数性故障诊断具有可行性和精确性。关键词：光伏并网逆变器；变分模态分解；小波包能量；海洋捕食者算法；长短时记忆网络；故障诊断中图分类号：TM464文献标志码：A文章编号：1003-8930（2023）02-0140-08DOI：10.19635/j.cnki.csu-epsa.001039ResearchonVMD-WPEandMPA-LSTMDiagnosticMethodsforParametricFaultsofPhotovoltaicGrid-connectedInvertersZHANGBide，YUHaining，LUORongqiu，ZHANGJin，FENGJing（SchoolofElectricalEngineeringandElectronicInformation，XihuaUniversity，Chengdu610039，China）Abstract:Aimedatproblemssuchasthedifficultyinextractingcharacteristicquantitiesandthelowdiagnosticaccura⁃cyofparametricfaultsofthree-phasephotovoltaic（PV）grid-connectedinverters，afaultdiagnosismethodwhichcom⁃binthewaveletpacketenergy（WPE）featureofvariationalmodaldecomposition（VMD）andthemarinepredatoralgo⁃rithm（MPA）tooptimizethelongshort-termmemory（LSTM）neuralnetworkisproposed.Thismethodusesthethree-phaselinevoltageoftheinverterastherawdataandtheminimumsampleentropyasthecriteriontooptimizethemodalnumberofVMD.Then，waveletpacketdecompositionisusedtoextracttheWPEofeachmodalcomponentofVMDasthefaultcharacteristicquantity.Finally，theMPAoptimizationLSTMnetworkhyperparameterisusedtoachievetheparametricidentificationoffaults.ThecomparativeanalysisresultsshowthefeasibilityandaccuracyoftheproposedmethodfortheparametricfaultdiagnosisofPVgrid-connectedinve...
