电工材料2023No.1田野等:基于PSO-XGBoost算法的多衰退特征锂离子电池SOH估计基于PSO-XGBoost算法的多衰退特征锂离子电池SOH估计田野a,闵锦涛b(三峡大学a.电气与新能源学院;b.计算机与信息学院,湖北宜昌443000)摘要:为了准确估计锂离子电池的健康状态(SOH),提出了一种基于粒子群算法与极限梯度提升算法相结合的方法。首先利用主成分分析法(PCA)对电池数据进行预处理,提取并组成最佳健康因子数据组;在此数据的基础上,运用XGBoost算法建立锂离子电池退化过程模型,利用同类已有电池历史数据进行训练,通过粒子群算法优化XGBoost算法中五个主要参数,构建基于PSO-XGBoost的SOH预测模型;最后采用美国国家航空航天局电池数据集进行分析验证,并与现有的预测方法对比。结果表明,该方法平均绝对误差为0.003922、均方根误差为0.005553、最大误差为0.02184,具有较高的预测精度。关键词:锂离子电池;健康状态;XGBoost;粒子群算法中图分类号:TM912DOI:10.16786/j.cnki.1671-8887.eem.2023.01.006SOHPredictionofLithiumIonBatteryWithMultipleDegradationCharacteristicsBasedonPSO-XGBoostAlgorithmTIANYea,MINJintaob(a.CollegeofElectricalEngineering&Newenergy;b.CollegeofComputerandInformationTechnology,ChinaThreeGorgesUniversity,HubeiYichang443000,China)Abstract:InordertoaccuratelyestimatethestateofHealth(SOH)oflithium-ionbatteriesunderthedoublecarbontarget,amethodbasedonparticleswarmoptimizationandlimitgradientliftingalgorithmisproposed.Firstly,thebatterydataispreprocessedbyprincipalcomponentanalysis(PCA),andthebesthealthfactordatagroupisextractedandformed.Onthebasisofthisdata,thedegradationprocessmodeloflithium-ionbatteryisestablishedbyusingXGboostalgorithm.Duringthetrainingprocess,thehistoricaldataofthesamekindofbatteriesareusedfortraining,andthefivemainparametersinXGboostalgorithmareoptimizedbyparticleswarmoptimizationalgorithmtobuildaSOHpredictionmodelbasedonPSO-XGboost.Finally,thebatterydatasetofNASAisusedforanalysisandverification,andthecomparisontestwiththeexistingpredictionmethodsshowsthattheaverageabsoluteerrorofthismethodis0.003922,therootmeansquareerroris0.005553,andthemaximumerroris0.02184,whichhashighpredictionaccuracy.Ke...