基于多任务集成学习的储能电池剩余使用寿命预测  

Multi-Task Ensemble Learning-Based Prediction of Remaining Useful Life of Energy-Storage Batteries

作　　者：王伟亮 刘会巧[2,3] 张天宇 阮鹏 徐劲 肖迁 WANG Weiliang;LIU Huiqiao;ZHANG Tianyu;RUAN Peng;XU Jin;XIAO Qian(State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210024,China;Zhonghuan Information College,Tianjin University of Technology,Tianjin 300380,China;Key Laboratory of Smart Grid of Ministry of Education(Tianjin University),Tianjin 300072,China;State Grid Tianjin Electric Power Company Economic and Technological Research Institute,Tianjin 300171,China;Pinggao Group Energy Storage Technology Co.,Ltd.,Tianjin 300300,China;Changchun Power Supply Company of State Grid Jilin Electric Power Co.,Ltd.,Changchun 130021,China)

机构地区：[1]国网江苏省电力有限公司,南京市210024 [2]天津理工大学中环信息学院,天津市300380 [3]智能电网教育部重点实验室(天津大学),天津市300072 [4]国网天津市电力公司经济技术研究院,天津市300171 [5]平高集团储能科技有限公司,天津市300300 [6]国网吉林省电力有限公司长春供电公司,长春市130021

出　　处：《电力建设》2024年第11期25-33,共9页Electric Power Construction

基　　金：国家自然科学基金项目(52107121);天津市自然科学基金多元投入重点项目(22JCZDJC00710);天津市企业科技特派员项目(23YDTPJC00090);天津大学自主创新基金项目(2024XHX-0028)。

摘　　要：“双碳”目标驱动下,电动汽车在交通能源转型中发挥关键作用,准确的剩余使用寿命(remaining useful life,RUL)预测可以指导电动汽车电池定期维护和降低事故风险。因此,提出了一个基于多任务集成学习的锂离子电池RUL预测模型,以实现行驶工况下RUL的准确预测。首先,通过增量容量-差值电压曲线,将健康因子量化为电导率损失、活性材料损失和锂离子损失;通过电化学阻抗谱,计算欧姆阻抗、电荷转移阻抗、固体电解质阻抗和Warburg阻抗。其次,基于多任务学习,分析了特征间相关性,保证了对特征的充分利用,降低了实验成本。然后,基于自适应鲁棒损失的改进轻量型梯度提升机,构建RUL预测模型,提高了预测准确率。最后,通过行驶工况下电池实验数据(振动工况:静置、X轴、Y轴和Z轴),验证所提模型有效性。结果表明:所提预测模型能够实现平均绝对误差小于1.4%、平均绝对百分比误差小于0.06%、均方根误差小于1.20%,所提预测模型有助于提高RUL预测准确率,保证电池稳定、安全运行。Driven by the goal of achieving carbon peak and neutrality,electric vehicles are crucial in the transformation of transportation energy.Thus,the accurate prediction of the remaining useful life(RUL)can be useful in periodic maintenance and reduce the risk of accidents.Therefore,this paper proposes a multi-task ensemble learning-based model for accurately predicting the RUL of lithium-ion batteries under driving conditions.First,an incremental capacity-differential voltage curve is used to quantify the loss of conductivity,active material,and lithium ions.Electrochemical impedance spectroscopy is used to calculate the ohmic,charge transfer,solid electrolyte,and Warburg impedances.Second,based on multi-task learning,the inter-feature correlation is analyzed to ensure full utilization of the features and reduce the experimental cost.Subsequently,based on the light gradient boosting machine improved by adaptive robust loss,an RUL prediction model is constructed,and it improves the prediction accuracy.Experimental data under driving conditions(vibration conditions:reference,X-axis,Y-axis,and Z-axis)were used to verify the effectiveness of the proposed model.The results show that the proposed prediction model can achieve a mean absolute error of less than 1.4%,a mean absolute percentage error of less than 0.06%,and a root mean square error of less than 1.20%.The proposed prediction model can improve RUL prediction accuracy and ensure stable and safe operation of the battery.

关 键 词：锂离子电池 剩余使用寿命 行驶工况 多任务学习 集成学习 

分 类 号：TM911[电气工程—电力电子与电力传动]