36720157ChineseJournalofScientificInstrumentVol.36No.7Jul.20152014-12ReceivedDate2014-12*61301205、51317040302*12111.1500802.014010、。。。3。。TP206+.3TH165+.3A510.40Afusionpredictionmethodoflithium-ionbatterycycle-lifeLiuYuefeng12ZhaoGuangquan1PengXiyuan11.DepartmentofAutomaticTestandControlHarbinInstituteofTechnologyHarbin150080China2.SchoolofInformationEngineeringInnerMongoliaUniversityofScience&TechnologyBaotou014010ChinaAbstractAccordingtotheproblemsoftraditionallithium-ionbatteryremainingusefullifeRULpredictionmethodbasedonparticlefiltersuchasexcessiverelianceonbatteryexperiencedegradationmodelandthesingleinputvariableofthemodelafusionRULestimationap-proachforlithium-ionbatteryisproposedbasedonrelevancevectormachineRVMparticlefilterPFandtheautoregressiveARmod-el.ThedegradationtrendofbatteryhistoricaldataisextractedbyRVMandthetrendequationisbuilttoreplacethebatteryexperiencedegra-dationmodelwhichisadoptedasthestatetransitionequationofthePFalgorithm.Long-termtrendpredictionvaluesoftheARmodelareusedtoreplacetherealvaluesandthentheobservationequationofthePFalgorithmisconstructed.ThreemethodsareintegratedtoestimatethebatteryRUL.Experimentalresultsshowthatthepredictionprecisionoffusionmethodishighandtheproposeddatadrivenapproachismorecommonbecauseitcanavoidbuildingthecomplexexperiencedegradationmodelbasedonbatteryfailuremechanism.Keywordslithiumionbatteryrelevancevectormachineparticlefilterautoregressivemodelfusionmethod1、。。。1。、、、batterymanagementsys-temBMSBMS2-3。GPSunmannedaerialvehiclesUAVs4-5。714636。LiuD.T.78FPGARVMremainingusefullifeRUL9-1011。。RUL。MiaoQ12PFOlivaresB.E.13PFSahaB14-16PFRUL。RUL17-1819、720RUL。ZhouJ.B.21RULLiuD.T.22ARRPFRULLiuD.T.23RUL。RULrel-evancevectormachineRVM、particlefilterPFARRUL。PFRVMSahaB24HuY25RVMPF。RVMtt。tRVMPF。RVMPFPFtRUL。ARPF。2RVM、PFARRVMPF2。AR。3。2.1RVMRVMTipping2001supportvectormachineSVMRVM26。RVMMercer。RVMSVM。xiyiMi=1xiyiMRVMyxw=∑Nn=1wnkxxn+w0ui=yxiw+εii=12…M1w=w0w1…wNNRVkxxnεi~N0σ2uiyi。wεi。αi。wα-1ipwi|αi~Nwi|0α-1i2pw|α=∏Ni=0Nwi|0α-1i3α=α0α1...αNΦ。ασαnewi=γi/μ2i4σ2new=u-Φμ2N-Σiγi5Σ=σ-2ΦTΦ+A-1μ=σ-2ΣΦTuγi=1-αiΣii。αi。αiwi。RVM26。2.2PFPF。PF27。2。。146436。PF。51xt=fxt-1vt6zt=hxtωt7txtztxtvtωt。2pxt|xt-16。3wit=wit-1pzt|xit84。。effectivesamplesizeESSESSt=∑ni=1wit2-19nwi=1/ni=12…nESStn。ESSt。PFTESSTESS=n/2ESSt<TESS。27。5xtpxt|z0t≈∑ni=1witδxt-xit10n10。2.3AR、PHMAR。ARxtxt=1xt-1+2xt-2+…+pxt-p+atp≠{011Φii=12…patt=0±1…σ2a。pARARp。ARxtARpp+2p、Φ1Φ2…Φpσ2a11AR。AR。ΦpppΦp。AkaikeinformationcriterionAIC12AICp=Nσ2a+2p12pNσ2ap。2.43PF6。。PF。RVM。RVMPFPF。RVM。ut~N∑Ni=1wnkxtxn+w0σ213RVMut。utPF。piti=12…nt=12…TnT。pit~N∑Ni=1wnkxtxn+w0σ214utRVM。14-16PFRVMPF。PFAR7PF。31。714651Fig.1Theschematicdiagramofthefusionframeworkforlithium-ionbatteryRULestimation33。NASA。3.1NASANASAPCoE2A·h11.5A4.2V22A2.5V3electrochemicalimpedancespec-troscopyEIS0.1~5Hz。endoflifeEoL70%。3Battery#05、Battery#06、Battery#07Battery#18。healthindicatorHI。EoL70%。2A·h1.38A·h。1.38A·hRUL。CALCE。1100mA·hCS2。CS2CS2_36、CS2_37、CS2_383RUL。31C4.2V0.05A2.7V。70%。3.2TkNCapacityCapoutkUCapacitykPFCapoutkRULprobabilitydensityfunctionPDFTCapacityRVMPFPFCapoutkURULPDF。1Capacity2TTT3TRVMPF4TAR5PFNEoLU6k=173RVMXik84ARZik9珓xik=πxk|xi0k-1y1k10珘wik11Neff=1ΣNi=1珘wik2Neff≥Nthresxi0k=珓xi0kwik=珘wik珘wlkKi=lxi0k=珓xKi0kwik=1/N12Capcoutk=ΣNi=1xi0kwik13k=k+17~13Capoutk14CapoutkEoLURULRUL=k15RULPDF。1466364NASAPCoECALCERULAR14PF7RVM。4.1PFNASAPCoEPF。18T=40、T=60、T=80EoLU=1.38A·hN=500R=0.0001Q=0.0001。T=80RUL2RUL1。2PFRULT=80Fig.2RULestimationresultswithproposedfusionprognosticalgorithmandstandardPFalgorithmT=801PFRULTable1RULestimationresultswithproposedfusionprognosticalgorithmandstandardPFalgorithmEoPRUL_predictionRUL_errorT=40763624PF773723T=6095355PF872713T=8098182PF899111TEoPendofpredictionRUL_predictionRUL_error15RUL_error=|RUL_prediction-RUL_true|15RUL_true。21T=4018EoL100RUL_true=100-T=60。EoP76RUL_prediction=76-T=36RUL_error=|36-60|=24。PFEoP77RUL_prediction=77-T=37RUL_er-ror=|37-60|=23。T=60RUL_true=100-T=40。EoP95RUL_prediction=95-T=35RUL_error=|35-40|=5。PFEoP87RUL_prediction=87-T=27RUL_error=|27-40|=13。T=80RUL_true=100-T=20。EoP98RUL_prediction=98-T=18RUL_error=|18-20|=2。PFEoP89RUL_prediction=89-T=9RUL_error=|9-20|=11。meanabsoluteerrorMAErootmeansquareerrorRMSE。MAERMSE1617MAE=1n∑ni=1|xi-珋xi|16RMSE=1n∑ni=1xi-珋xi槡217nxii珋xi。PFMAERMSE2。2PFRULTable2RULestimationresultserrorwithproposedfusionprognosticalgorithmandstandardPFalgorithmMAERMSE1014PF16172MAERMSEPFPFPFPF。714674.2RVMCALCERVM。51。37EoL142×5T=60×5、T=80×5、T=100×5、T=120×5EoLU=0.79A·hN=500R=0.0001Q=0.0001。2RUL33。3RVMRULT=100Fig.3RULestimationresultswithproposedfusionprognosticalgorithmandRVMalgorithmT=1003RVMRULTable3RULestimationresultswithproposedfusionprognosticalgorithmandRVMalgorithmEoPRUL_predictionRUL_errorT=6087×527×555×5RVM80×520×562×5T=80117×537×525×5RVM104×524×538×5T=100145×545×53×5RVM151×551×59×5T=120147×527×55×5RVM130×510×512×5RVMMAERMSE4。4RVMRULTable4RULestimationresultserrorwithproposedfusionprognosticalgorithmandRVMalgorithmMAERMSE10×514×5RVM16×517×533RVMRVMRVM。4MAERMSERVM。5RVM、PFARRVMPFPFARPFPFRVM。PF。RVMPF。1.J.20153611-16.L