基于GA-BO-LSTM的电解电容剩余寿命预测

打开文本图片集

引用格式：，，.基于GA-BO-LSTM的电解电容剩余寿命预测[J].现代电子技术，2025，48（20）：81-86.

关键词：剩余寿命预测；遗传算法；贝叶斯优化；长短期记忆网络；超参数优化；调优区间；分层抽样中图分类号：TN86-34；TM535 文献标识码：A 文章编号：1004-373X（2025）20-0081-06

RULpredictionofelectrolyticcapacitorsbasedonGA-BO-LSTM

LIUXinyi，LIXiaobo，SHI Shangxian （SchoolofUrbanRailTransportation，ShanghaiUniversityofEngineeringScience，Shanghai2O162O，China）

Abstract：Theremainingusefullife（RUL）predictionofsingleproductisofgreatsignificanceforthestableandreliable operationofthesystem.Inorder toimprovetheaccuracyof theRULpredictionofelectrolyticcapacitorsandresolve the problemsofsuddensystemfailurescausedbylowpredictionaccuracy，amethodofRULpredictionthatcombiningGenetic algorithm（GA）and Bayesianoptimization（BO）todeterminethehyper-parametersof thelongshort-termmemory（LSTM）network isproposed.Hermite interpolationisusedforthedatapreprocesing.GAandBOareusedtoperformglobaloptimization respectively on three hyperparameters of the LSTM model： the initial learning rate，the L2 regularization coefficient，and the numberof hidden layercells.Thetwoobtained hyperparametersrepresenting thesame meaning wereservedas boundaryvalues toconstructanovelparameterstuning interval.Latinhypercubesampling（LHS）wasemployedforthestratifiedsampling，and therootmeansquareerorand meanabsoluteerror wereused todeterminetheoptimalnumberoflayersas6.Thealgorithmwas validatedbasedontheacelerateddegradationtestingdataofelectrolyticcapacitorsfromNASA.Theresultsshowthatthe proposed algorithm canreduetheerorbyatleast38.57%comparedwithLSTM，GA-LSTM，andBO-LSTM，demonstrating significantlyhigherprediction accuracy.

Keywords：remaininguseful lifeprediction;geneticalgorithm;Bayesianoptimization;longshort-termmemory;hyperparameter optimization; optimization interval;stratified sampling

0 引言

产品的剩余寿命（RemainingUsefulLife，RUL）预测可提前识别潜在故障，提升设备的可靠性和安全性1。（剩余6467字）