第50卷第4期2023年4月Vol.50,No.4Apr.2023湖南大学学报(自然科学版)JournalofHunanUniversity(NaturalSciences)基于TCN编码的锂离子电池SOH估计方法周航1,程泽2†,弓清瑞2,刘旭2(1.天津大学建筑设计规划研究总院有限公司,天津300073;2.天津大学电气自动化与信息工程学院,天津300072)摘要:为了能够准确可靠地估计锂离子电池的健康状态(StateofHealth,SOH),提出一种基于时序卷积网络(TemporalConvolutionalNetwork,TCN)的数据驱动模型来建立电池充电曲线与SOH之间的映射关系.TCN是一种由多层因果卷积组成的神经网络,它能够对电池充电曲线上的采样点序列进行编码,通过编码得到的编码向量会更易于与SOH建立映射关系.实验结果表明所提基于TCN的SOH估计模型具有较高的估计精度,对不同种类的电池也有良好的适应能力.关键词:锂离子电池;充电曲线;健康状态;时序卷积网络;神经网络中图分类号:TM912.1文献标志码:ASOHEstimationMethodofLithium-ionBatteryBasedonTCNEncodingZHOUHang1,CHENGZe2†,GONGQingrui2,LIUXu2(1.TianjinUniversityResearchInstituteofArchitecturalDesignandUrbanPlanningCo.,Ltd,Tianjin300073,China;2.SchoolofElectricalandInformationEngineering,TianjinUniversity,Tianjin300072,China)Abstract:Thestateofhealth(SOH)ofalithium-ionbatteryreflectstheagingdegreeofLithium-ionthebattery.Whenthebatteryischargedinconstantcurrent-constantvoltagemode,thechargingcurveswithdifferentagingdegreesarealsodifferent.Basedonthisfact,thispaperproposesadata-drivenmodelbasedonatemporalconvolutionalnetwork(TCN)toestablishthemappingrelationshipbetweenthechargingcurveandSOH.TCNisanovelneuralnetworkcomposedofmulti-layercausalconvolution,whichcanencodethesequenceofsamplingpointsonthechargingcurve.TheexperimentprovesthattheencodingvectoriseasiertoestablishthemappingrelationshipwithSOH.TheexperimentalresultsshowthattheproposedSOHestimationmodelhashighestimationaccuracyandgoodadaptabilitytodifferenttypesofbatteries.Keywords:lithium-ionbatteries;chargecurves;stateofhealth(SOH);temporalconvolutionalnetwork(TCN);neuralnetworks锂离子电池是一种较为清洁的储能装置,有体积小、成本低、能量密度高、循环寿命长等优点,...