设为首页 联系我们 加入收藏

当前位置: 网站首页 期刊分类目录 2025第3期 正文

基于 CNN-Transformer 模型的锂离子电池 SOC 预测

作者:时间:2025-07-04点击数:



全文下载: 3.pdf


文章编号:1672-6987(2025)03-0136-07;DOI:10. 16351/j. 1672-6987. 2025. 03. 018


陈柳成 1 ,吕 岩 2 ,张 伟 1* (1. 青岛科技大学 自动化与电子工程学院,山东 青岛 266061;2. 山东科技大学 教务处,山东 青岛 266590)


要:准确地预测锂离子电池荷电状态(state of charge,SOC)对电动汽车安全行驶和电池管理等具有重要的指导意义。为提高锂离子电池 SOC 预测精度,提出一种卷积神经网络CNN 与 Transformer 模型结合的预测方法。该方法充分利用 Transformer 模型中多头自注意力机制的优势,挖掘 SOC 与锂电池放电数据的复杂的映射关系以及 SOC 序列存在的长期依赖性,实现并行运算。同时结合 CNN 网络的特征提取优势,提取对当前时刻的 SOC 影响较大的深层放电特征。在动态工况数据集下的锂电池 SOC 预测实验表明,本工作预测方法在不同温度下预测的平均误差仅为 0. 57%,具有较高的预测精度和应用价值。


关键词:锂离子电池;荷电状态; Transformer 模型; CNN 卷积神经网络


中图分类号:TM 911 文献标志码:A


引用格式:陈柳成,吕岩,张伟 . 基于 CNN-Transformer 模型的锂离子电池 SOC 预测[J]. 青 岛科技大学学报(自然科学版),2025,46(3):136-142.


CHEN Liucheng, LYU Yan, ZHANG Wei. Prediction of lithium-ion battery SOC based on CNN-Transformer model[J]. Journal of Qingdao University of Science and Technology

Natural Science Edition),2025,46(3):136-142.


Prediction of Lithium-Ion Battery SOC Based on CNN-Transformer Model


CHEN Liucheng1 , LYU Yan2 ,ZHANG Wei1 (1. College of Automation and Electronic Engineering, Qingdao University of Science and Technology, Qingdao 266061,

China;2. Academic Affairs Office, Shandong University of Science and Technology, Qingdao 266590, China)


Abstract:Predicting state of charge (SOC) of lithium-ion battery accurately plays an impor tant role in ensuring the safe operation of electric vehicle and battery management. In order to improve the SOC prediction accuracy, a prediction method combining convolution neural net work (CNN) and Transformer model is proposed. This method makes full use of the advan tages of Transformer model encoder and decoder structure to dig out the complex mapping rela tionship between SOC and lithium-ion battery discharge data, as well as the long-term depen dence of SOC sequence, and realizing parallel computation. At the same time, the CNN net

work extract the historical discharge features that have a great influence on the SOC at the cur rent moment, reducing the calculation cost and memory capacity. The experimental results of SOC prediction for lithium-ion battery under dynamic working condition data set show that the average prediction error of the proposed prediction method is 0. 57% under different tempera tures, which has high prediction accuracy and application value.


Key words:lithium-ion battery; state of charge; Transformer model; convolutional neural net work


收稿日期:2024-07-21

基金项目:青岛市关键技术攻关及产业化示范类项目(23-1-2-qljh-3-gx);青岛市社会科学规划研究项目(QDSKL2401119).

作者简介:陈柳成(1997—),女,硕士研究生 . 通信联系人 .


Copyright © 2011-2017 青岛科技大学学报 (自然科学版)