基于GA-LSTM神經(jīng)網(wǎng)絡(luò)的充電樁故障診斷

打開文本圖片集

中圖分類號：TN919-34；TP183 文獻標識碼：A 文章編號：1004-373X（2025）16-0097-08

DOI：10.16652/j.issn.1004-373x.2025.16.016

Charging pile fault diagnosis based on GA-LSTM neural network

ZHOU Jin， GAO Tian，WANG Qiang， YIN Zhangcheng， ZHU Jinrong （Collegeof InformationEngineering，Yangzhou University，Yangzhou 225ooo，China）

Abstract：Thechargingdataofelectricvehiclechargingfacilitiesincludesvoltage，current，temperature，power，etc.allof whicharetimeseriesdata，andthedatahasthecharacteristicsofinfluencingtheprevioustimeandasociatingthenexttime. Thelongshort-termmemory（LSTM）neuralnetworkisusedtominethecorelationfeaturesinthedatavolume，andthefeature modelbetweentheworkingdataandthefaultisestablishedtoconductthefaultdiagnosisofchargingpile.Duetotheproblems ofoverfitingandgetingstuck inlocaloptimainLSTMneuralnetworks，ageneticalgorithm（GA）isproposed tooptimizethe LSTMneuralnetwork forfaultdiagnosisofcharging piles.GAisusedtosearch fortheoptimal solutionbysimulating the biologicalevolutionprocessindividualswithhighfiessareselectedforeproduction，andmutationoperationsareintroducedt graduallyoptimizethecombinationofhyperparameters，soastoimprovetheperformanceandeficiencyoftheLSTMmodel.In comparison with the experimental resultsof LSTMneural network，the RMSEvalueand MPAEvalueof GA-LSTM neural network prediction results are decreased by 56.7% and 60.3% ，respectively，and the accuracy rate of fault diagnosis is increased by 3.2% .Therefore，GA-LSTMneural network can beusedasa dep learning technologyforthe fault diagnosis of charging pile.

Keywords：charging pile;dataprediction;faultdiagnosis;geneticalgorithm;long short-termmemoryneural network; normalization processing

0 引言

隨著電動汽車銷量逐年增長，電動汽車充電樁作為配套產(chǎn)品也得到快速發(fā)展，其功能越來越完善，智能化程度也越來越高。（剩余9283字）