基于卷积注意力机制的锂离子电池剩余使用寿命预测研究  

Research on prediction of remaining useful life of lithium⁃ion batteries based on convolutional attention mechanism

作　　者：史云 汪立伟[1] 公岷[1] SHI Yun;WANG Li-wei;GONG Min(School of Electronics and Information,Southwest Minzu University,Chengdu 610041,China)

机构地区：[1]西南民族大学电子信息学院,四川成都610041

出　　处：《西南民族大学学报（自然科学版）》2024年第3期336-346,共11页Journal of Southwest Minzu University(Natural Science Edition)

基　　金：西南民族大学中央高校基本科研业务费专项资金资助(2022NYXXS096)。

摘　　要：锂离子电池剩余使用寿命的准确估计和预测对于锂离子电池电源管理系统有着重要的意义.一方面能够提高实际电路系统的可靠性,同时能够延长电池的使用寿命.锂离子电池的退化过程表现出复杂、非线性特征且伴随容量再生现象,导致传统预测模型对锂离子电池剩余使用寿命预测(Remaining Useful Life,RUL)准确性低.为进一步提升锂离子电池RUL的预测精度,首先深入分析电池历史退化数据,构建一种新的基于卷积注意力机制深度学习框架.通过对锂离子电池老化循环过程的研究,选取容量数据作为健康因子(Health Indicator,HI),利用卷积神经网络(Convolu⁃tional Neural Network,CNN)的卷积和池化操作挖掘数据内在信息,降低数据复杂度,提取电池数据的时序特征.并将特征数据输出到构建的注意力机制(Attention Mechanism,AM)深度网络,捕捉全局时序数据的位置信息和分析时序数据中内部信息的关系,以此获得准确的RUL预测.最后使用公开的NASA和CALCE数据集上进行验证,并对比其他几种预测模型,结果表明所提出模型的具有较高预测精度和泛化适应能力.The accurate estimation and prediction of the remaining useful life of lithium⁃ion battery is of great significance for lithium⁃ion battery power management system.On the one hand,it can improve the reliability of the actual circuit system,and prolong the useful life of the battery.The degradation process of lithium⁃ion batteries shows complex and nonlinear characteris⁃tics and is accompanied by capacity regeneration,which leads to the low accuracy of traditional prediction models for Remaining Useful Life(RUL)prediction of lithium⁃ion batteries.In order to further improve the prediction accuracy of lithium⁃ion battery RUL,this paper firstly analyzed the historical battery degradation data and constructed a new deep learning framework based on convolutional attention mechanism.Through the research on the aging cycle process of lithium⁃ion battery,the capacity data were selected as the Health Indicator(HI),and the convolution and pooling operation of Convolutional Neural Network(CNN)was used to mine the internal information of the data.The data complexity was reduced and the timing characteristics of battery data were extracted.The feature data were output to the constructed Attention Mechanism(AM)deep network to capture the location information of the global time series data and analyze the relationship between the internal information in the time series data,so as to obtain accurate RUL prediction.Finally,the proposed model was verified on the public NASA and CALCE datasets,and compared with several other prediction models.The results showed that the proposed model had higher prediction accuracy and generalization adaptability.

关 键 词：锂电池 剩余使用寿命 健康因子 卷积注意力机制 注意力神经网络 

分 类 号：TM912[电气工程—电力电子与电力传动]