汽车工程 ›› 2021, Vol. 43 ›› Issue (9): 1291-1299.doi: 10.19562/j.chinasae.qcgc.2021.09.004

• • 上一篇    下一篇

基于实车数据的电动汽车电池健康状态预测

胡杰(),朱雪玲,何陈,杨光宇   

  1. 1.武汉理工大学,现代汽车零部件技术湖北省重点实验室,武汉 430070
    2.武汉理工大学,汽车零部件技术湖北省协同创新中心,武汉 430070
    3.新能源与智能网联车湖北工程技术研究中心,武汉 430070
  • 收稿日期:2021-03-11 修回日期:2021-05-09 出版日期:2021-09-25 发布日期:2021-09-26
  • 通讯作者: 胡杰 E-mail:auto_hj@163.com
  • 基金资助:
    湖北省科技重大专项(2020AAA001)

Prediction on Battery State of Health of Electric Vehicles Based on Real Vehicle Data

Jie Hu(),Xueling Zhu,Chen He,Guangyu Yang   

  1. 1.Wuhan University of Technology,Hubei Key Laboratory of Modern Auto Parts Technology,Wuhan 430070
    2.Wuhan University of Technology,Auto Parts Technology Hubei Collaborative Innovation Center,Wuhan 430070
    3.Hubei Technology Research Center of New Energy and Intelligent Connected Vehicle Engineering,Wuhan 430070
  • Received:2021-03-11 Revised:2021-05-09 Online:2021-09-25 Published:2021-09-26
  • Contact: Jie Hu E-mail:auto_hj@163.com

摘要:

鉴于现有电动汽车电池健康状态(SOH)预测方案多基于条件有限实验室的实验数据,且存在单指标预测精度低等问题,基于实车运行数据分析并提取电池健康状态因子,以电池容量、内阻和单体一致性为特征,构建机器学习模型,实现电池SOH多指标的准确预测;针对实车数据区间不完整、片段间隔大等问题,提出自适应状态估计法;利用非支配排序遗传算法(NSGA?II)进行精度与效率的多目标优化,获得最佳电压区间,提高电池容量的变区间估计精度。结果表明,该方法可有效实现基于实车数据的电池SOH准确预测,采用5-fold交叉验证计算测试集最大平均绝对误差小于2%。

关键词: 电动汽车, SOH预测, 机器学习

Abstract:

In view of that most of the existing battery state of health (SOH) prediction schemes are based on the experimental data obtained in the laboratory with limited conditions, and the poor accuracy of single indicator prediction, a machine learning model is constructed based on the analysis on real vehicle operating data and the extraction of battery health state factors, with the battery capacity, internal resistance and cell consistency as its features, to achieve accurate prediction of battery SOH with multiple indicators. Aiming at the problems of the incomplete interval of real vehicle data and the large interval of segments, an adaptive state estimation method is proposed. Non-dominated sorting genetic algorithm (NSGA?II) is used to conduct a multi-objective (accuracy and efficiency) optimization, with the optimal voltage interval obtained and the accuracy of variable interval estimation of battery capacity enhanced. The results show that the method proposed can effectively achieve the accurate prediction of battery SOH based on real vehicle data, with the maximum mean absolute error of test set less than 2% when using 5?fold cross validation.

Key words: electric vehicle, SOH prediction, machine learning