




中圖分類號:TM911.4 文獻標志碼:A DOI:10.20104/j.cnki.1674-6546.20240429
LifePrediction of Proton Exchange Membrane Fuel Cell Based onISAO-CNN-GRU
Xiong Jianyu1,Kuang Yazhoul,Peng Yiqiangl,2
(1.ScholofAutomobileand Transportation,Xihua University,Chengdu 610o39;2.Vehicle MeasurementControlandSafety KeyLaboratoryofSichuanProvince,Chengdu61Oo39;3.ProvincialEngineering Research CenterforNewEnergy Vehicle Intelligent Control and Simulation Test Technology of Sichuan, Chengdu 610039)
【Abstract]To predict the Remaining Useful Life(RUL)of Proton Exchange Membrane Fuel Cell (PEMFC)precisely,the paper proposes a method for predicting the RUL based onneural network optimized by Improved Snow Ablation Optimizer (ISAO).Firstlytheoriginaldataarepreprocessedbyusing Pautacriterionandwavelets,thenthePearson’scorrelation coeficients areused toselect parameters which have strong corelation with voltageas input variables.ISAOisused to optimize hyperparametersof Convolutional Neural Network-GatedRecurent Unit(CNN-GRU) model.Thenthe CNN-GRU model isusedtopredicttheoutputvoltageof the PEMFC.Testresults show that whenthetraining setratio is 30%,the mean absoluteerroris 1.6mV ,theroot mean square erroris 2.2mV ,therelativeerroris 0.41% ,and theR-squared of themethod is 99.20%,whicharethe bestresults theof six models.Compared with the Sparow Search Algorithm (SSA),Snow Ablation Optimizer (SAO)and Whale Optimization Algorithm(WOA),the ISAO hasfasteroptimization speed and beterresult,proving that the prediction model and the improved algorithm are effective.
Keywords:Proton Exchange Membrane Fuel Cell(PEMFC),Remaining Useful Life (RUL). SnowAblation Optimizer (SAO),Gauss-Cauchy mutation
【引用格式】熊健宇,匡亞洲,彭憶強.基于ISAO-CNN-GRU的質(zhì)子交換膜燃料電池壽命預測[J].汽車工程師,2025(7): 36-43. XIONGJY,KUANGYZ,PENGYQ.LifePredictionof Proton Exchange MembraneFuel CellBasedon ISAOCNN-GRU[J]. Automotive Engineer, 2025(7): 36-43.
*基金項目:四川省科技廳重大科技項目(2019ZDZX0002);四川省區(qū)域創(chuàng)新合作項目(2020YFQ0037);四川省重點研發(fā)計劃項目(2021YFG0071)。
1前言
2 基本原理
質(zhì)子交換膜燃料電池(ProtonExchangeMembraneFuelCell,PEMFC)在許多領域得到了廣泛應用。雖然其具有能量轉(zhuǎn)換效率高、工作噪聲小、無污染等多種優(yōu)點,但目前仍面臨維護成本高、性能衰減快、耐久性不足等問題2。精準預測PEMFC的剩余使用壽命(RemainingUsefulLife,RUL)是確保PEMFC得到及時維護,進而延長其使用壽命的關鍵因素。
目前,RUL的預測方法包括模型驅(qū)動法、數(shù)據(jù)驅(qū)動法和混合驅(qū)動法。模型驅(qū)動法通過構建數(shù)學模型表征燃料電池的退化特性,主要包括根據(jù)燃料電池退化的物理及化學原理構建的機理模型,以及通過挖掘電池內(nèi)部各種參數(shù)的數(shù)學關系構建的經(jīng)驗模型[4-5,預測算法主要有卡爾曼濾波及其變體6-8、粒子濾波等。數(shù)據(jù)驅(qū)動法直接運用燃料電池老化過程中的數(shù)據(jù)即可實現(xiàn)準確預測。文獻[10]提出了一種多灰色組合結合反向傳播(BackPropagation,BP)神經(jīng)網(wǎng)絡的預測模型,運用氫燃料電池汽車的實時數(shù)據(jù)進行訓練,預測結果與實際結果相近。文獻[11]提出一種支持向量機(SupportVectorMachine,SVM)與經(jīng)驗模型結合的燃料電池壽命混合預測模型,預測誤差相比SVM模型平均下降了 80% 。文獻[12]在基于長短期記憶(Long Short-TermMemory,LSTM)網(wǎng)絡的預測方法的基礎上增加卷積神經(jīng)網(wǎng)絡(ConvolutionalNeuralNetworks,CNN),加快了訓練和預測速度,相對誤差僅為0.03% 。文獻[13]提出基于門控循環(huán)單元(GatedRecurrentUnit,GRU)的燃料電池RUL預測方法,與BP和LSTM網(wǎng)絡相比,預測精度得到大幅提升。……