基于差分电压和Elman神经网络的锂离子电池RUL预测方法  

RUL prediction of lithium-ion battery based on differential voltage and Elman neural network

作　　者：李练兵[1] 李思佳 李洁[1] 孙坤 王正平 杨海跃 高冰 杨少波 LI Lianbing;LI Sijia;LI Jie;SUN Kun;WANG Zhengping;YANG Haiyue;GAO Bing;YANG Shaobo(State Key Laboratory of Reliability and Intelligence of Electrical Equipment,Hebei University of Technology,300130,Tianjin,China;School of Artificial Intelligence,Hebei University of Technology,300130,Tianjin,China;Hengshui power supply company of State Grid Hebei Electric Power Co.Ltd.,Hengshui 053099,Hebei,China;State Grid Hebei Electric Power Co.,Ltd.Electric Power Research Institute,Shijiazhuang 050021,Hebei,China)

机构地区：[1]河北工业大学省部共建电工装备可靠性与智能化国家重点实验室,天津300130 [2]河北工业大学人工智能与数据科学学院,天津300130 [3]国网河北省电力有限公司衡水供电公司,河北衡水053099 [4]国网河北省电力有限公司电力科学研究院,河北石家庄050021

出　　处：《储能科学与技术》2021年第6期2373-2384,共12页Energy Storage Science and Technology

基　　金：河北省重点研发计划项目(20312102D)。

摘　　要：锂离子电池剩余使用寿命(remaining useful life,RUL)预测对电池的使用维护极为重要,提出一种基于差分电压和Elman神经网络预测锂离子电池RUL的方法。首先,根据美国国家航天航空局(National Aeronautics and Space Administration,NASA)的锂离子电池数据集,分析电池差分电压曲线和充放电曲线,提取电池容量退化特征量;其次,通过Pearson法分析特征量之间的相关性,将充电差分电压曲线初始拐点值、放电差分电压曲线峰值、放电时间、静置时间作为电池RUL预测的间接健康因子;最后,建立以上述间接健康因子为输入,电池容量为输出的Elman神经网络,进行锂离子电池的RUL预测。基于不同间接健康因子和不同神经网络的四种电池容量预测对比实验表明,在间接健康因子中加入充电差分电压曲线初始拐点值和放电差分电压曲线峰值可以提高电池寿命预测精度,Elman神经网络可准确预测电池容量。基于不同循环次数预测电池RUL,预测的平均均方根误差为1.55%。The prediction of the Remaining Useful Life(RUL)of lithium-ion battery has significant value for the study of the usage and maintenance of batteries.A method for RUL prediction of lithium-ion batteries based on differential voltage and Elman neural network is proposed.Firstly,based on the National Aeronautics and Space Administration’(NASA)lithium-ion battery data set,the differential voltage curve and charge-discharge curve of the battery are analyzed,and the characteristic quantity of battery capacity degradation is extracted.Secondly,the correlation between characteristic parameters is studied by Pearson method.The inflexion point of charging differential voltage curve,peak value of discharging differential voltage curve,as well as the discharging time and resting time are determined as indirect health factors of battery RUL prediction.Finally,an Elman neural network employed the inputs of above indirect health factors and the output of battery capacity is established to predict RUL of lithium-ion batteries.The comparative experiments of four kinds of battery capacity prediction based on different indirect health factors and different neural networks show that the prediction accuracy of battery life can be improved by adding the initial inflection point of charging differential voltage curve and the peak value of discharge differential voltage curve into the indirect health factors,and the Elman neural network can accurately predict the battery capacity.The mean root mean square error(RMSE)of the prediction of battery RUL based on different cycle times is 1.55%.

关 键 词：锂离子电池 RUL 差分电压 ELMAN神经网络 相关系数 

分 类 号：TM911.3[电气工程—电力电子与电力传动]