储能科学与技术 ›› 2021, Vol. 10 ›› Issue (6): 2326-2333.doi: 10.19799/j.cnki.2095-4239.2021.0099

• 储能测试与评价 • 上一篇    下一篇

基于注意力改进BiGRU的锂离子电池健康状态估计

王凡(), 史永胜(), 刘博亲, 左玉洁, 符政, ALI Jamsher   

  1. 陕西科技大学电气与控制工程学院,陕西 西安 710021
  • 收稿日期:2021-03-15 修回日期:2021-05-13 出版日期:2021-11-05 发布日期:2021-11-03
  • 作者简介:王凡(1997—),男,硕士研究生,研究方向为锂离子电池健康管理系统,E-mail:270929211@qq.com|史永胜,教授,研究方向为特种开关电源与新型电源技术研究,E-mail:35743980@qq.com
  • 基金资助:
    国家自然科学基金项目(61871259);陕西省科技厅工业科技攻关计划项目(2019GY-175);陕西省教育厅专项科研计划项目(18JK0111)

Health state estimation of lithium-ion batteries based on attention augmented BiGRU

Fan WANG(), Yongsheng SHI(), Boqin LIU, Yujie ZUO, Zheng FU, Jamsher ALI   

  1. School of Electrical and Control Engineering, Shaanxi University of Science&Technology, Xi'an 710021, Shaanxi, China
  • Received:2021-03-15 Revised:2021-05-13 Online:2021-11-05 Published:2021-11-03

摘要:

锂离子电池的健康状态(state of health, SOH)是电池管理系统的核心问题,对其精确的评估能够保障电池的安全可靠运行。然而在实际应用中,容量较难直接测得,导致SOH估算困难。为了获得准确的SOH,本文提出一种基于注意力改进双向门控循环单元(BiGRU)的锂离子电池SOH估计方法。首先提取电池充放电曲线中的电压、电流与阻抗等参数,通过自编码器(auto encoder, AE)对其降维,提取特征量并减少数据间的冗余性。其次,引入注意力机制(attention mechanism, AM)对输入变量分配权重,突出对SOH估计起到关键作用的特征量。最后,利用BiGRU学习输入变量与容量之间的映射关系,捕获容量衰减下的长期依赖性。在不同充电倍率的电池数据集上的结果表明,该方法对不同类型电池的SOH皆可以实现高精度估计,均方根误差在1.1%以下。

关键词: 锂离子电池, 健康状态, 自编码器, 注意力机制, 双向门控循环神经网络

Abstract:

Lithium-ion batteries' state of health (SOH) is the central concern of a battery management system. An accurate evaluation of SOH can ensure that batteries operate safely and reliably. However, in practice, it is difficult to measure capacity, which makes SOH estimation difficult directly. To obtain accurate SOH, this paper proposes an attentional improved bidirectional gated recurrent unit (BiGRU)-based SOH estimation method for lithium-ion batteries. Firstly, parameters such as voltage, current, and impedance are extracted from the charge-discharge curve of the battery, and the auto encoder reduces the dimensions to extract the features and reduce the redundancy between the data. Secondly, an attention mechanism is introduced to assign weight to input variables and highlight the characteristic quantities that play a key role in SOH estimation. Finally, the BiGRU is used to learn the mapping relationship between input variables and capacity and capture long-term dependence under capacity decay. The results of the University of Maryland battery datasets with different charging rates show that the proposed method can estimate SOH with high precision for different types of batteries, with a root mean square error of less than 1.1%.

Key words: lithium-ion battery, state of health(SOH), attention mechanism, auto encoder, bidirectional gated recurrent unit

中图分类号: