基于TransformerEncoder-DR1DCNN的锂离子电池RUL预测  

TransformerEncoder-DR1DCNN-based prediction of remaining useful life for lithium-ion batteries

作　　者：王浩[1] 李亚[1] 王海瑞[1] 朱贵富[2] WANG Hao;LI Ya;WANG Hairui;ZHU Guifu(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China;Information Construction Management Center,Kunming University of Science and Technology,Kunming 650504,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650504 [2]昆明理工大学信息化建设管理中心,云南昆明650504

出　　处：《陕西理工大学学报(自然科学版)》2025年第2期54-63,共10页Journal of Shaanxi University of Technology:Natural Science Edition

基　　金：国家自然科学基金项目(61863016)。

摘　　要：针对锂离子电池的剩余使用寿命(RUL)预测,提出了一种基于Transformer编码器层(TransformerEncoder)与深度残差一维卷积神经网络(DR1DCNN)相结合的预测方法。首先提取容量数据作为直接健康因子,并对容量数据进行归一化处理以消除数据量纲影响;接着使用滑动时间窗口机制构建容量时序序列数据,并划分训练集和测试集;然后采用TransformerEncoder捕捉容量时序序列全局各个位置之间的相关性以及序列长距离的依赖关系,使用DR1DCNN提取局部相邻数据间的关联关系。最后采用不同预测起点的多步预测方式以检验模型的有效性。以NASA公开的数据集进行实验,两组电池的均方根误差不超过2%,平均绝对误差不超过1.4%。并通过与其他文献的实验结果进行对比,验证了所提方法能够提前多步预测锂离子电池的RUL,以起到早期预警作用。For the prediction of the Remaining Useful Life(RUL)of lithium-ion batteries,a method based on the combination of Transformer Encoder layers(TransformerEncoder)and Deep Residual 1D Convolutional Neural Network(DR1DCNN)is proposed.First,capacity data is extracted as a direct health indicator and normalized to eliminate the influence of data dimensions.Next,a sliding time window mechanism is used to construct capacity time series data,which is divided into training and testing sets.The TransformerEncoder is then employed to capture the correlations between different positions in the global capacity time series and the long-range dependencies within the sequence,while the DR1DCNN is utilized to extract the relationships between locally adjacent data points.Finally,a multi-step prediction method with different starting points is adopted to validate the effectiveness of the model.Experiments conducted on the publicly available NASA dataset show that the root mean square error for two sets of batteries is less than 2%,and the mean absolute error is less than 1.4%.By comparing with experimental results from other literature,it is verified that the proposed method can predict the RUL of lithium-ion batteries multiple steps in advance,providing early warning capabilities.

关 键 词：Transformer编码器层 深度残差一维卷积神经网络 多步预测 锂离子电池 剩余使用寿命预测 

分 类 号：TM912[电气工程—电力电子与电力传动]