储能科学与技术

• 储能科学与技术 •    

基于变分模态分解与特征增强的锂电池健康状态估计方法

张吴哲1(), 蔡志端2,3(), 吴成傲1, 郑炜1, 童嘉阳2   

  1. 1.湖州师范学院工学院,浙江省 湖州 313000
    2.湖州学院智能制造学院,浙江省 湖州 313000
    3.湖州市绿色能源材料与电池梯次利用重点实验室,浙江省 湖州 313000
  • 收稿日期:2025-01-08 修回日期:2025-01-13
  • 通讯作者: 蔡志端 E-mail:1035176174@qq.com;caizhiduan@zjhzu.edu.cn
  • 作者简介:张吴哲(2001—),男,研究生,学生,研究方向:锂电池性能状态估计方法研究,E-mail:1035176174@qq.com
  • 基金资助:
    浙江省湖州市基础公益研究计划项目(2022gz02)

Lithium Battery State of Health Estimation Method Based on Variational Mode Decomposition and Feature Enhancement

Wuzhe ZHANG1(), Zhiduan CAI2,3(), Chengao WU1, Wei ZHENG1, Jiayang TONG2   

  1. 1.School of Engineering, Huzhou University, Huzhou, Zhejiang 313000, China
    2.School of Intelligent Manufacturing, Huzhou University, Huzhou, Zhejiang 313000, China
    3.Huzhou Key Laboratory of Green Energy Materials and Battery Cascade Utilization, Huzhou, Zhejiang 313000, China
  • Received:2025-01-08 Revised:2025-01-13
  • Contact: Zhiduan CAI E-mail:1035176174@qq.com;caizhiduan@zjhzu.edu.cn

摘要:

锂电池因其高能量密度、生命周期长等优点,已广泛应用于电动汽车、储能系统和便携设备等领域。然而,锂电池在使用过程中不可避免地会出现容量再生现象,这一现象会影响电池健康状态的评估精度。针对这一问题,本文提出了一种结合变分模态分解和生成对抗网络的电池健康状态估计方法,以降低容量再生现象对电池健康状态评估精度的影响。具体而言,首先通过变分模态分解对锂电池在放电过程中反映容量再生现象的等压降放电时间特征量进行多尺度分解,并根据皮尔逊相关系数分析,分离出主退化趋势分量与局部波动分量,以更好地表征电池健康状态。主退化趋势分量反映了电池整体退化的趋势,而局部波动分量则反映了容量再生现象引起的局部变化特性。接着,为了进一步增强局部波动分量中与容量再生现象相关的特征,对其进行傅里叶变换,提取其中反映容量再生现象的中低频段分量。针对这些频段的分量,使用生成对抗网络进行数据生成,将生成的数据与变分模态分解后的多特征结合,构建新的多特征集。然后,采用支持向量机算法对新的多特征集进行训练,实现对锂电池健康状态的准确估计。最后,基于NASA数据集和CALCE数据集进行验证实验。实验结果表明,所提出的方法相比于传统方法,均方根误差均能控制在4.5%以内,能够有效降低容量再生现象对电池健康评估精度的影响。

关键词: 容量再生, 局部波动分量, 傅里叶变换, 支持向量机

Abstract:

Because of its high energy density and long life cycle,lithium batteries have been widely used in electric vehicles, energy storage systems and portable devices. However, capacity regeneration is inevitable during the use of lithium batteries, which will affect the assessment accuracy of battery health. In order to reduce the influence of capacity regeneration on battery health assessment accuracy,a new method combining variational mode decomposition and generative adversarial network was proposed.Firstly, the variational mode decomposition was used to decompose the iso-pressure drop discharge time characteristic,which reflects the capacity regeneration phenomenon in the discharge process of lithium battery, and the main degradation trend component and local fluctuation component were separated according to Pearson correlation coefficient analysis,so as to better characterize the health state of the battery.The main degradation trend component reflects the overall degradation trend of the battery. The local fluctuation component reflects the local variation characteristics caused by capacity regeneration. Secondly, in order to further enhance the relevant characteristics of local wave components and capacity regeneration phenomenon,Fourier transform is carried out to extract the middle and low frequency components reflecting the capacity regeneration phenomenon.For the components of these frequency bands, the generated adversarial network is used to generate data, and the generated data is combined with the multi-feature after variational mode decomposition to construct a new multi-feature set. Then, support vector machine algorithm is used to train the new multi-feature set to achieve accurate estimation of the health state of the lithium battery.Finally, validation experiments are conducted based on NASA data sets and CALCE data sets. The experimental results show that compared with traditional methods, the RMS error of the proposed method can be controlled within 4.5%, which can effectively reduce the influence of capacity regeneration on the battery health assessment accuracy.

Key words: Capacity regeneration, Local fluctuation component, Fourier Transform, Support Vector Machine

中图分类号: