电工技术学报  2023, Vol. 38 Issue (11): 3073-3083    DOI: 10.19595/j.cnki.1000-6753.tces.220187
电能存储与应用 |
基于改进DBSCAN的退役动力电池分选方法
刘征宇, 郭乐凯, 孟辉, 张政, 刘项
合肥工业大学机械工程学院 合肥 230009
A Sorting Method of Retired Power Battery Based on Improved DBSCAN
Liu Zhengyu, Guo Lekai, Meng Hui, Zhang Zheng, Liu Xiang
School of Mechanical Engineering Hefei University of Technology Hefei 230009 China
全文: PDF (3951 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 退役动力电池梯次利用时电池单体不一致性对于分选后电池组性能具有重要影响,高效的分选方法能够降低电池不一致性,提高电池组的使用性能和安全性。针对目前常用的电池检测系统采样频率较低等问题,首先使用自适应分段拟合方法对充放电数据进行拟合,从充放电曲线中提取表征电池不一致性的动态特征电压上升高度(VR)、电压下降深度(DVF),并与容量、开路电压静态特征结合构成分选特征向量;然后提出一种基于核密度估计的DBSCAN算法(KDE-DBSCAN),通过核密度估计自适应确定聚类算法参数,对特征聚类实现电池的分选;最后通过实验验证该分选方法的有效性。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
刘征宇
郭乐凯
孟辉
张政
刘项
关键词 退役动力电池梯次利用分选DBSCAN自适应    
Abstract:The rapid growth of electric vehicle sales has led to an increase in the number of decommissioned lithium power batteries. Testing and echelon utilization sorting for retired batteries can maximize the service life of batteries. Traditional battery echelon utilization sorting methods usually only consider external static parameters and neglect the difference between battery dynamic characteristics, so the sorting accuracy is limited. Recently, some sorting methods based on dynamic characteristic curves have been proposed, but most of them have problems such as complex feature extraction and poor optimization of algorithm parameters. This paper proposes a DBSCAN algorithm based on kernel density estimation (KDE-DBSCAN). The parameters of the clustering algorithm are adaptively determined by kernel density estimation, and the combination of static features and dynamic features is clustered to achieve battery sorting.
Firstly, this study adopts sixth-order polynomial to adaptively fit the charge and discharge curve which is collected by the detetion system in sections. Secondly, extract the HVR (height of voltage rise) and DVF (depth of voltage falls) as the dynamic characteristics during charging and discharging, and the available capacity and open circuit voltage after charging as the static features. Thirdly, the extracted dynamic and static features of the battery are combined to form a four-dimensional feature vector and normalized. Finally, the feature sample points are clustered by DBSCAN clustering algorithm to finish the sorting of retired batteries after the parameters of the method are adaptively set by kernel density estimation. In this method, kernel density estimation solves the problem of DBSCAN parameter optimization, reduces human interference, and fully improves the accuracy and robustness of clustering.
The sorting and grouping results of KDE-DBSCAN, DBSCAN and K-means clustering algorithms are compared and verified by charge discharge experiments. The results show that in the cell characteristics experiment, the mean value of battery voltage sorted by KDE-DBSCAN is larger and the variance is smaller compared with DBSCAN and K-means clustering algorithms. The average and variance of the polarization voltage of the battery pack based on KDE-DBSCAN sorting are significantly lower than those of the other two methods. The smaller mean value indicates that the overall polarization of the battery pack is smaller. The smaller variance indicates that the polarization fluctuation between single cells in the battery pack is smaller, and the battery pack is more inconsistent. In the battery module cycle life comparison experiment, from the 20th cycle to the 50th cycle, the decrease in discharge time of the KDE-DBSCAN sorted battery is significantly lower than that of batteries sorted by K-means clustering and DBSCAN clustering. After 50 cycles, the capacity of the former decreases by about 10.2% compared with the initial capacity. For the battery packs sorted by K-means and DBSCAN clustering, the numbers are 20.8% and 18.2% respectively.
The following conclusions can be drawn from the experimental analysis: 1) The adaptive subsection fitting method can effectively solve the problem of low sampling frequency of battery detection system and improve the accuracy of feature extraction. 2) The combination of dynamic and static characteristics can better characterize the inconsistency of retired batteries. 3) Kernel density estimation is introduced to determine the parameters of DBSCAN algorithm, which reduces the instability of manual parameter adjustment and improves the accuracy and robustness of clustering.
Key wordsRetired power battery    echelon utilization    sorting    DBSCAN    adaptive   
收稿日期: 2022-02-10     
PACS: TM912  
基金资助:安徽省自然科学基金(1808085MF200)和工业和信息化部民用飞机专项科研项目(MJ-2017-D-26)资助
通讯作者: 郭乐凯 男,1998年生,硕士,研究方向为动力电池梯次利用分选、制造大数据分析等。E-mail:1043244508@qq.com   
作者简介: 刘征宇 男,1979年生,副教授,硕士生导师,研究方向为电池能量系统建模与控制、智能制造与工业物联网、制造大数据分析等。E-mail:liuzhengyu@hfut.edu.cn
引用本文:   
刘征宇, 郭乐凯, 孟辉, 张政, 刘项. 基于改进DBSCAN的退役动力电池分选方法[J]. 电工技术学报, 2023, 38(11): 3073-3083. Liu Zhengyu, Guo Lekai, Meng Hui, Zhang Zheng, Liu Xiang. A Sorting Method of Retired Power Battery Based on Improved DBSCAN. Transactions of China Electrotechnical Society, 2023, 38(11): 3073-3083.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.220187          https://dgjsxb.ces-transaction.com/CN/Y2023/V38/I11/3073