800W.Lietal./Engineering5(2019)795–802Fig.4.Standarddifferenceofbatterymodule.categories(i.e.,twomodulespurchasedfromthemanufacturerwiththesamespeci?cations,oneSVC-clusteredbatterymodule,andonek-means-clusteredbatterymoduleproducedfromthegroupingofcells).TheexperimentalsetupisshowninFig.5.Aircoolingwassuppliedfromthebottomforthemodulesofeachcat-egory.Thetemperaturewasobservedevery5minoverthecycleasthemodulewascharged–dischargedatthesamerate.Fig.6clearlyshowsthatthebatterymodulescorrespondingtoCategory3(theSVC-clusteredbatterymodule)presentedthebestperformance,withamaximumobservedtemperatureof32°C.Bycontrast,themaximumobservedtemperaturesoftheotherbatterymoduleswerehigher,at40°CforCategory1(manufacturer),36°CforCategory2(manufacturer)and35°CforCategory4(k-means-clusteredbatterymodule).AstheSVC-clusteredbatterymoduleunderwenttheleastheating,itisexpectedtohavealongerlife-cyclethanthemodulesintheothercategories.Aplausiblerea-sonforthisresultisthattheselectionofcellswithsimilarperfor-mancethatwasmadewhenproducingthemoduleresultedinanFig.5.Experimentalsetupfortheveri?cationoftheproducedbatterymodules.W.Lietal./Engineering5(2019)795–802801Fig.6.Temperaturevariationofthebatterymodulesatsixdifferentpositionsinacharging–dischargingcyclefromfourcategories:(a)Category1(manufacturer);(b)Category2(manufacturer);(c)Category3(SVC-clusteredbatterymodule);(d)Category4(k-means-clusteredbatterymodule).equalizedtemperaturedistributionwithinthemodule,whichcon-sequentlyloweredtheriseintemperatureincomparisonwiththemodulesintheothercategories.5.ConclusionsToachieveuniformityandequalizationoftheLi-ioncellsusedinabatterymoduleforNEVs,wecombinedexperimentalandnumericalmethodstoconductacomprehensiveinvestigationontheclusteringofbatterycellswithsimilarperformanceinordertodesignabatterymodulewithbetterelectrochemicalperfor-mance.Charging–dischargingtestswereperformedon48cells.Clusteringalgorithmswerethenemployedtoconductaclusteringanalysisonthetwokindsofbatterymodules(aSVC-clusteredbat-terymoduleandak-means-clusteredbatterymodule).Theperfor-mancesofthebatterymodulescreatedusingclusteringalgorithmswerecomparedwiththeperformancesofthetwomodulespur-chasedfromamanufacturer.TheSVC-clusteredbatterymoduleexhibitedthebestperformance,withamaximumobservedtem-peratureof32°C.Bycontrast,themaximumobservedtempera-turesoftheotherbatterymoduleswerehigher,at40°CforCategory1(manufacturer),36°CforCategory2(manufacturer),and35°CforCategory4(k-means-clusteredbatterymodule).Aplausiblereasonforthis?ndingisthattheselectionofcellswithsimilarperformanceduringtheproductionofthemoduleresultedinanequalizedtemperaturedistributionwithinthemodule,whichconsequentlyloweredtheriseintemperatureincomparisonwiththemodulesintheothercategories.Thek-meansclusteringalgorithmperformancemayvarydependingonthedataused.However,fortheSVCalgorithm,ifthedataaregiven,theclusteringresultsareonlyaffectedbytheSVCparametersettings.Furthermore,sinceSVCavoidsexplicitcal-culationsinthehigh-dimensionalfeaturespace,itiseffectiveforlargedatasets.Itcaneasilybeappliedinindustrialcontextsinwhichtheelectricvehiclescomprisehundredsofpacks.Inordertominimizebatterymanufacturingdefects,thepro-cessingtechnologyandassemblylevelcanbeimproved;alterna-tively,theabilitytodetectdefectscanbeimproved.However,manufacturingdefectsdoexist.Althoughtheproposedapproachmayappeartobeoverlylengthyforincorporationbeforethedesignstage,itisworthnotingthatanalternativeapplicationoftheproposedmethodcouldbeforbatteryrecycling.Sincebatteriescontainchemicalsubstancesandheavymetals,theirdisposalcancauseenvironmentalpollutionandawasteofresources.However,oldbatteriesstillhavevariouslevelsofcapacitythatcanbeusedinotherareas.Futureworkcanfocusonconductinglarge-scaletest-ingoncellsinordertodesignalargerbatterymodule,aswellasonperformingexperimentalveri?cationontheperformanceofprobabilisticmethods[43,44],extrememachinelearningmethods[45,46],andarti?cial-intelligence-basedmethods[47–50].AcknowledgementsThisworkwassupportedbytheNationalNaturalScienceFoun-dationofChina(51675196and51721092)andtheprogramforHUSTAcademicFrontierYouthTeam(2017QYTD04).Theauthorsacknowledgethegrant(DMETKF2018019)fromtheStateKeyLabofDigitalManufacturingEquipmentandTechnology,HuazhongUniversityofScienceandTechnology;theSailingTalentProgramandtheGuangdongUniversityYouthInnovationTalentProject(2016KQNCX053)supportedbytheDepartmentofEducationofGuangdongProvince;andtheShantouUniversityScienti?cResearchFundedProject(NTF16002).CompliancewithethicsguidelinesWeiLi,SiqiChen,XiongbinPeng,MiXiao,LiangGao,AkhilGarg,andNengshengBaodeclarethattheyhavenocon?ictofinterestor?nancialcon?ictstodisclose.References[1]ChoiJW,AurbachD.Promiseandrealityofpost-lithium-ionbatterieswithhighenergydensities.NatRevMater2016;1(4):16013.[2]WenF,LinC,JiangJC,WangZG.Anewevaluationmethodtotheconsistencyoflithium-ionbatteriesinelectricvehicles.In:Proceedingsof2012Asia-Paci?c802W.Lietal./Engineering5(2019)795–802PowerandEnergyEngineeringConference;2012Mar27–29;Shanghai,China;2012.[3]MohantyD,HockadayE,LiJ,HensleyDK,DanielC,WoodIIIDL.Effectofelectrodemanufacturingdefectsonelectrochemicalperformanceoflithium-ionbatteries:cognizanceofthebatteryfailuresources.JPowerSources2016;312:70–9.[4]HongL,LiLS,Chen-WiegartYK,WangJJ,XiangK,GanLY,etal.Two-dimensionallithiumdiffusionbehaviorandprobablehybridphasetransformationkineticsinolivinelithiumironphosphate.NatCommun2017;8(1):1194.[5]FangKZ,ChenS,MuDB,WuBR,WuF.Investigationofnickel-metalhydridebatterysortingbasedonchargingthermalbehavior.JPowerSources2013;224:120–4.[6]ShiW,HuXS,JinC,JiangJC,ZhangYR,YipT.Effectsofimbalancedcurrentsonlarge-formatLiFePO4/graphitebatteriessystemsconnectedinparallel.JPowerSources2016;313:198–204.[7]YangNX,ZhangXW,ShangBB,LiGJ.Unbalanceddischargingandagingduetotemperaturedifferencesamongthecellsinalithium-ionbatterypackwithparallelcombination.JPowerSources2016;306:733–41.[8]BrandMJ,HofmannMH,SteinhardtM,SchusterSF,JossenA.Currentdistributionwithinparallel-connectedbatterycells.JPowerSources2016;334:202–12.[9]DubarryM,DevieA,LiawBY.Cell-balancingcurrentsinparallelstringsofabatterysystem.JPowerSources2016;321:36–46.[10]WeiX,ZhuB.TheresearchofvehiclepowerLi-ionbatterypackbalancingmethod.In:Proceedingsofthe9thInternationalConferenceonElectronicMeasurement&Instruments;2009Aug16–19;Beijing,China;2009.[11]ParkSH,ParkKB,KimHS,MoonGW,YounMJ.Single-magneticcell-to-cellchargeequalizationconverterwithreducednumberoftransformerwindings.IEEETransPowerElectr2012;27(6):2900–11.[12]SunFC,XiongR.Anoveldual-scalecellstate-of-chargeestimationapproachforseries-connectedbatterypackusedinelectricvehicles.JPowerSources2015;274:582–94.[13]MooreSW,SchneiderPJ.Areviewofcellequalizationmethodsforlithiumionandlithiumpolymerbatterysystems.SAETechnicalPaper.Warrendale:SocietyofAutomotiveEngineers,Inc.;2001.ReportNo.:2001-01-0959.[14]PeiL,ZhuCB,WangTS,LuRG,ChanCC.Onlinepeakpowerpredictionbasedonaparameterandstateestimatorforlithium-ionbatteriesinelectricvehicles.Energy2014;66:766–78.[15]AnFQ,HuangJ,WangCY,LiZ,ZhangJB,WangS,etal.Cellsortingforparallellithium-ionbatterysystems:evaluationbasedonanelectriccircuitmodel.JEnergyStorage2016;6:195–203.[16]Gallardo-LozanoJ,Romero-CadavalE,Milanes-MonteroMI,Guerrero-MartinezMA.Batteryequalizationactivemethods.JPowerSources2014;246:934–49.[17]KimJ,ShinJ,ChunC,ChoBH.Stablecon?gurationofaLi-ionseriesbatterypackbasedonascreeningprocessforimprovedvoltage/SOCbalancing.IEEETransPowerElectr2012;27(1):411–24.[18]KimJ,ChoBH.Screeningprocess-basedmodelingofthemulti-cellbatterystringinseriesandparallelconnectionsforhighaccuracystate-of-chargeestimation.Energy2013;57:581–99.[19]KimCH,KimMY,ParkHS,MoonGW.Amodularizedtwo-stagechargeequalizerwithcellselectionswitchesforseries-connectedlithium-ionbatterystringinanHEV.IEEETransPowerElectr2012;27(8):3764–74.[20]LiXY,WangTS,PeiL,ZhuCB,XuBL.Acomparativestudyofsortingmethodsforlithium-ionbatteries.In:Proceedingsof2014IEEEConferenceandExpoTransportationElectri?cationAsia-Paci?c;2014Aug31–Sep3;Beijing,China;2014.[21]WangQ,ChengXZ,WangJ.Anewalgorithmforafasttestingandsortingsystemappliedtobatteryclustering.In:Proceedingsofthe6thInternationalConferenceonCleanElectricalPower;2017Jun27–29;SantaMargheritaLigure,Italy.Piscataway:IEEE;2017.p.397–402.[22]EnamiN,MoghadamRA.Energybasedclusteringselforganizingmapprotocolforextendingwirelesssensornetworkslifetimeandcoverage.CanJMultimedWirelNetw2010;1(4):42–54.[23]RaspaP,FrinconiL,ManciniA,CavallettiM,LonghiS,FulimeniL,etal.SelectionoflithiumcellsforEVbatterypackusingself-organizingmaps.AutomotSafEnergyTechnol2011;2:32–9.[24]PiaoCH,WangZG,CaoJ,ZhangW,LuS.Lithium-ionbatterycell-balancingalgorithmforbatterymanagementsystembasedonreal-timeoutlierdetection.MathProblEng2015;2015:168529.[25]MaY,DuanP,SunYS,ChenH.Equalizationoflithium-ionbatterypackbasedonfuzzylogiccontrolinelectricvehicle.IEEETransIndElectron2018;65(8):6762–71.[26]HeF,ShenWX,SongQ,KapoorA,HonneryD,DayawansaD.ClusteringLiFePO4cellsforbatterypackbasedonneuralnetworkinEVs.In:Proceedingsof2014IEEEConferenceandExpoTransportationElectri?cationAsia-Paci?c;2014Aug31–Sep3;Beijing,China;2014.[27]LiXY,SongK,WeiG,LuRG,ZhuCB.AnovelgroupingmethodforlithiumironphosphatebatteriesbasedonafractionaljointKalman?lterandanewmodi?edk-meansclusteringalgorithm.Energies2015;8(8):7703–28.[28]YangYX,GaoMY,HeZW,WangCS.Arobustbatterygroupingmethodbasedonacharacteristicdistributionmodel.Energies2017;10(7):1035.[29]LeeKM,ChungYC,SungCH,KangB.ActivecellbalancingofLi-ionbatteriesusingLCseriesresonantcircuit.IEEETransIndElectron2015;62(9):5491–501.[30]ShangYL,ZhangCH,CuiNX,GuerreroJM.Acell-to-cellbatteryequalizerwithzero-currentswitchingandzero-voltagegapbasedonquasi-resonantLCconverterandboostconverter.IEEETransPowerElectr2015;30(7):3731–47.[31]LeeKM,LeeSW,ChoiYG,KangB.ActivebalancingofLi-ionbatterycellsusingtransformerasenergycarrier.IEEETransIndElectron2017;64(2):1251–7.[32]EinhornM,RoesslerW,FleigJ.ImprovedperformanceofseriallyconnectedLi-ionbatterieswithactivecellbalancinginelectricvehicles.IEEETransVehTechnol2011;60(6):2448–57.[33]ZhengYJ,LuLG,HanXB,LiJQ,OuyangMG.LiFePO4batterypackcapacityestimationforelectricvehiclesbasedonchargingcellvoltagecurvetransformation.JPowerSources2013;226:33–41.[34]ZhongL,ZhangCB,HeY,ChenZH.Amethodfortheestimationofthebatterypackstateofchargebasedonin-packcellsuniformityanalysis.ApplEnergy2014;113:558–64.[35]ZhengYJ,OuyangMG,LuLG,LiJQ,HanXB.XuLF.On-lineequalizationforlithium-ionbatterypacksbasedonchargingcellvoltages:part1.Equalizationbasedonremainingchargingcapacityestimation.JPowerSources2014;247:676–86.[36]SamadiMF,SaifM.NonlinearmodelpredictivecontrolforcellbalancinginLi-ionbatterypacks.In:Proceedingsof2014AmericanControlConference;2014Jun4-6;Portland,OR,USA.Piscataway:IEEE;2014.p.2924–9.[37]XuR,WunschDJr.Surveyofclusteringalgorithms.IEEETransNeuralNetw2005;16(3):645–78.[38]TzortzisG,LikasA.TheMinMaxk-meansclusteringalgorithm.PatternRecognit2014;47(7):2505–16.[39]MacQueenJ.Somemethodsforclassi?cationandanalysisofmultivariateobservations.In:ProceedingsoftheFifthBerkeleySymposiumonMathematicalStatisticsandProbability;1965Jun21–Jul18,1965Dec27–1966Jan7;Berkeley,CA,USA.Berkeley:UniversityofCaliforniaPress;1967.p.281–97.[40]HungCH,ChiouHM,YangWN.Candidategroupssearchfork-harmonicmeansdataclustering.ApplMathModel2013;37(24):10123–8.[41]Ben-HurA,HornD,SiegelmannHT,VapnikV.Supportvectorclustering.JMachLearnRes2001;2(12):125–37.[42]JunS,ParkSS,JangDS.Documentclusteringmethodusingdimensionreductionandsupportvectorclusteringtoovercomesparseness.ExpertSystAppl2014;41(7):3204–12.[43]GargA,HazraB,ZhuH,WenYP.Asimpli?edprobabilisticanalysisofwatercontentandwiltinginsoilvegetatedwithnon-cropspecies.Catena2019;175:123–31.[44]GargA,BordoloiS,MondalS,NiJJ,SreedeepS.Investigationofmechanicalfactorofsoilreinforcedwithfourtypesof?bers:anintegratedexperimentalandextremelearningmachineapproach.JNatFibers2018;2018:1–15.[45]GargA,ShankhwarK,JiangD,VijayaraghavanV,PandaBN,PandaSS.Anevolutionaryframeworkinmodellingofmulti-outputcharacteristicsofthebonedrillingprocess.NeuralComputAppl2018;29(11):1233–41.[46]GargA,PengXB,LeMLP,PareekK,ChinCMM.Designandanalysisofcapacitymodelsforlithium-ionbattery.Measurement2018;120:114–20.[47]ZhouWH,TanF,YuenKV.Modelupdatinganduncertaintyanalysisforcreepbehaviorofsoftsoil.ComputGeotech2018;100:135–43.[48]TanF,ZhouWH,YuenKV.Effectofloadingdurationonuncertaintyincreepanalysisofclay.IntJNumerAnalMethodsGeomech2018;42(11):1235–54.[49]PandaB,LeiteM,BiswalBB,NiuXD,GargA.Experimentalandnumericalmodellingofmechanicalpropertiesof3Dprintedhoneycombstructures.Measurement2018;116:495–506.[50]PandaBN,GargA,ShankhwarK.Empiricalinvestigationofenvironmentalcharacteristicof3-Dadditivemanufacturingprocessbasedonslicethicknessandpartorientation.Measurement2016;86:293–300.Engineering 2 (2016) xxx–xxxContents lists available at ScienceDirect
Engineering
ResearchClean Energy—Article电动汽车锂电池模块设计中相似性能电池聚类的综合方法李伟a，陈思琦b，彭雄斌b，肖蜜a，高亮a,*，Akhil Garg b，包能胜bab State Key Lab of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074, China Key Lab of Mechatronic Systems Intelligent Integration Technology, Ministry of Education, Shantou University, Shantou 515063, Chinaa r t i c l e i n f oArticle history:Received 28 August 2018Revised 25 October 2018Accepted 3 June 2019Available online 10 July 2019摘要新能源汽车的核心组成部分为能量存储系统，该系统由多个锂电池模块组成，为车辆传动系统提供主要动力。然而模块中的单体电池由于生产制造的缺陷，在性能上往往表现出差异。这些差异的存在会导致电池模块的不完全充放电以及温度分布的不均匀，进而导致循环寿命和电池容量随着时间的推移而降低。为解决这一问题，本工作采用实验和数值方法对性能相似的电池进行了全面的聚类研究，从而得到了电化学性能更好的电池模块。首先通过模块拆解实验来测量电池性能参数，并基于k-均值聚类与支持向量聚类算法设计电池模块，每个模块均由12块电池组成。然后在风冷条件下测量一定时间内电池模块的实际温升，验证聚类设计的效果。研究发现第三类（支持向量聚类）电池模块的性能最佳，充放电最高观测温度为32 ℃。相比之下，其他电池模块的最高温度值要更高：第一类（厂家原装）电池模块为40 ℃，第二类（厂家原装）电池模块为36 ℃，以及第四类（k-均值聚类）电池模块为35 ℃。? 2019 THE AUTHORS. Published by Elsevier LTD on behalf of Chinese Academy of Engineering and Higher Education Press Limited Company This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).关键词聚类算法电池模块均衡电动汽车1.引言电池模块作为新能源汽车能量存储系统的主要组成部分，可用来替代传统汽车的燃油（汽油/柴油）系统并对环境无污染，因而获得越来越广泛的关注[1,2]。电池模块由单体电池通过串并联的方式组成，其性能参数包括能量密度、容量和比功率等。为给电动汽车提供足够的动力，实际情况下通常将小型电池模块通过串并联的方式组成规模更大的电池模块（也被称为电池组）。电动汽车的性能范围取决于其内部电池模块的性能，而电池模块的性能又取决于单体电池的性能及电池的串并联配置方式。理想的电池模块应遵循均匀性与均衡性的标准，然而，这些标准尚未得到很好的实施。* Corresponding author. E-mail address: gaoliang@mail.hust.edu.cn (L. Gao). 在大规模制造电池和将电池组装成模块的过程中，由于生产环境不确定性而产生的偏差不可避免[3]。这些偏差包括电极材料性能的差异、工艺条件的变动以及加工机器精度误差导致的电池几何尺寸的变化等[4]，这些不确定性会造成电池模块的缺陷，如表面划伤、极片箔暴露以及裂纹。同时电池模块在串并联组装过程中的缺陷会导致模块整体的性能变化，进而影响单体电池的性能参数（即容量和电压）。一段时间后，这些累积的差异会造成电池温度的不均匀分布、模块中部分电池的不完全充放电以及电池实际容量的降低[5–7]。因此在电池模块设计制造过程中需遵守均匀性与均衡性标准，以有效地防止诸如过热、热失控等情况，从而提高电池模块的寿命[8–12]。Author name et al. / Engineering 2(2016) xxx–xxx883目前研究者已经提出一些电池分类方法来解决电池差异性带来的问题[13–15]。Gallardo-Lozano等[16]汇总分析了多种电池主动均衡系统的方法，同时总结出电池分类的最佳方法是使用开关电容及双层开关电容。Kim等[17]提出了一种筛选方法（容量筛选和内阻筛选），用以提高锂电池系列电池模块的实用性，并在后续研究中进一步提出一种串并联布置的多电池串通用建模方法[18]。Kim等[19]提出了一种带选择开关的模块化两级电荷均衡器。这种筛选方式的优势在于其可以广泛应用于混合动力汽车的大数量锂电池筛选过程中。此外，文献[20]提出了五种电池分类方法，并在容量、交流内阻、电化学阻抗谱、电压曲线、动力参数以及热性能参数等方面对五种方法进行对比分析，结果表明基于动态特性的低频电池阻抗法是电池分类的最佳方法。以前的研究[21–36]主要集中于同类电池的选择与分类，得出的结果表明：筛选后的电池在容量、电压和温度方面，较之未筛选的电池拥有更佳的一致性。但目前针对这些研究结果进行实验验证的工作较少。因此，本文提出实验和数值分析相结合的方法，对性能相近电池的聚类进行了全面的研究，并设计了一种拥有更高电化学性能的电池模块。图1介绍电池聚类分析及电池模块性能验证的流程。首先对48块锂电池进行充放电实验，测量其容量、电压与温度。然后基于k-均值聚类与支持向量聚类算法将电池进行分组并组装成电池模块。最后将本研究中设计的电池模块与从生产商购买的电池模块做性能上的对比分析。2. 数据测量实验本节介绍了测量48块锂电池数据（容量、电压和温度）的充放电实验，这48块电池由电池组拆解得到，如图2所示。电池模块拆解流程可分为以下四步：第一步，分析电池模块，获得基本信息，如容量、电池数目以及电池间连接方式。第二步，拆开电池模块外壳后，立即识别模块的输出端口。该步需特别注意，避免电池模块中任何正、负极端子发生接触。第三步，先打断电池间的串联连接。该步通过破坏电池间连接，将电池模块分成更小的电池单元，这样做的目的是确保拆解过程中的安全。第四步，将拆分成的电池单元进一步分解成单体电池。电池模块分解完成后，开始在电池测试系统上进行单体电池充放电测试，如图2所示。电池测试系统主要包括电池测试设备、数据采集系统、单体锂电池等。电图1. 电池模块设计与制造的综合流程。
电动汽车锂电池模块设计中相似性能电池聚类的综合方法 - 图文
