云端数据驱动的锂电池故障无监督学习早期预警

打开文本图片集

Unsupervised learning early warning of lithium battery failure driven by cloud data

ZHOU Zhengyi，YANG Lin*1，MENG Yizhen1，LIHuaijin1，LU Feng2，LIU Zhisheng1， LI Yang2，WUWeikun2

（1.SchoolofechanicalEngineering，hanghaiJiaoTongUniversity，o240，China； 2.ShanghaiQiyuanGreenPowerTechnologyCo.，Ltd.，20oool，China）

Abstract：An unsupervised learning early warning method was proposed based on voltage consistency to warn earlythelithiumbateryfaultsincloudbaterymanagementtechnology.Thevoltagecharacteristicsintheeffctive chargingcyclewereextracted with measuring thedegreeofvoltage consistencybyusingaminimumneighborhood radiuswhichachieved a singlecluster number for DBSCAN （density-based spatial clustering of applications with noise）；Aparameterwithdimension-onewasdefined toimprove thealgorithmgeneralizationabilitytotheactual working conditions；The hyperparameterssuchasalarm thresholdswere selected through orthogonal experiment. Theactualfault cases were verifiedandanalyzed.Theresultsshowthat forthe baterysystemswith thelowstate of charge（SOC）faults，thesinglebatteryundervoltagefaults，andthesingleconsistencyfaults，thismethodenables earlywarningmore than50days inadvance，withanaccuracyrateof 96.7% ，and can locate the cells ofsubsequently developsfaults.Therefore，earlywarningof ithium-battry-systemfailures isrealized throughunsupervisedlearning.

Keywords：electricvehicle；lithium-ionbattery（LiB）；batterymanagement；clouddata； unsupervised learning； fault warning；minimum neighborhood radius

锂离子电池（lithium-ionbattery，LiB）凭借能量密度高、循环寿命长和自放电率低等优点[是目前电动汽车电源的最佳选择。（剩余12875字）