汽车工程 ›› 2024, Vol. 46 ›› Issue (3): 489-497.doi: 10.19562/j.chinasae.qcgc.2024.03.013

• • 上一篇    

基于相似性优化模型样本的实车锂电池健康状态分析

刘万里1,李子涵2,3,梁宏毅1,陈吉清2,3(),莫丙达1   

  1. 1.广汽本田汽车有限公司, 广州 510700
    2.华南理工大学机械与汽车工程学院, 广州 510641
    3.华南理工大学, 广东省汽车工程重点实验室, 广州 510641
  • 收稿日期:2023-08-01 修回日期:2023-09-10 出版日期:2024-03-25 发布日期:2024-03-18
  • 通讯作者: 陈吉清 E-mail:chjq@scut.edu.cn

Real Vehicle Battery Health State Estimation Based on Similarity Optimization Model Samples

Wanli Liu1,Zihan Li2,3,Hongyi Liang1,Jiqing Chen2,3(),Bingda Mo1   

  1. 1.Guangqi Honda Automobile Company Limited,Guangzhou  510700
    2.School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou  510641
    3.South China University of Technology,Guangdong Provincial Automobile Engineering Key Laboratory,Guangzhou  510641
  • Received:2023-08-01 Revised:2023-09-10 Online:2024-03-25 Published:2024-03-18
  • Contact: Jiqing Chen E-mail:chjq@scut.edu.cn

摘要:

为更好地解决电动汽车动力电池健康状态(SOH)在线估计问题,减少实车采集数据中的冗余样本,改善运行工况不稳定导致的特征丢失,提升实车电池SOH估计的精度,提出一种基于增量容量分析方法(ICA)提取特征和动态时间规整(DTW)优化特征样本的SOH估计方法。首先对实车电池充电循环数据应用增量容量分析提取电池IC曲线,以曲线峰高度等形状特征作为健康因子。采用动态时间规整作相似性判据,基于IC曲线形状计算电池充电循环样本的相似度,保留与基准充电循环相似的充电循环数据,优化训练样本,最后采用全连接神经网络(MLP)模型进行SOH估计。以实车运行电池数据进行对比实验,结果表明该方法可明显改善训练样本质量,提升电池SOH估计精度。

关键词: 锂离子电池, 健康状态估计, 容量估计, 相似样本

Abstract:

In order to better solve the online estimation problem of electric vehicle power battery health state (SOH), reduce redundant samples in real-vehicle data collection, improve feature loss caused by unstable operating conditions, and enhance the accuracy of SOH estimation of real-vehicle batteries, a SOH estimation method based on incremental capacity analysis (ICA) method to extract features and dynamic time regularization (DTW) to optimize feature samples is proposed. Firstly, incremental capacity analysis is applied to extract the battery IC curve from the real battery charging cycle data, and shape features such as curve peak height are used as health factors. Then, dynamic time regularization is used as the similarity criterion to calculate the similarity of the battery charge cycle samples based on the IC curve shape, and the charge cycle data similar to the baseline charge cycle is retained to optimize the training samples. Finally, the fully connected neural network (MLP) model is used for SOH estimation. Comparative tests are conducted with real vehicle running battery data, and the results show that the method can significantly improve the training sample quality and enhance the battery SOH estimation accuracy.

Key words: lithium-ion battery, state-of-health estimation, capacity estimation, similar samples