基于改进的在线迁移学习算法的电池高压故障预警

doi:10.19799/j.cnki.2095-4239.2023.0119

摘要/Abstract

摘要：

锂电池作为一种电化学设备，在发生故障前的特征较为复杂，难以分析，并且生产环境中故障样本数量较少，正负样本比例严重不平衡。针对以上问题，本工作提出基于改进的在线迁移学习算法的电池高压故障预警。首先，引入下采样技术，解决样本不均衡问题，从而降低计算资源的使用。在电池高压故障预警场景下，设计分段下采样策略，使得算法模型在故障发生前能学习到更多细微的特征。其次，提出基于分批增量学习的在线迁移学习方法(homogeneous online transfer learning under incremental training，HomOTL-UIT)，源域中训练的离线分类器需要在合适的时间进行更新，以此来适应目标域中不断变化的数据分布，解决数据分布偏移和在线迁移学习退化为在线学习的问题。分批处理降低多次训练带来的计算资源的开销，通过增量学习，不断从目标域中学习，从而不断提高离线分类器的准确度。然后，设计一种滑动窗口下的F₁-score评分方法，解决模型权重缓慢失衡问题，从而提高模型的准确性。最后，通过储能集装箱的运行数据验证所提出方法的有效性和准确性，在正负样本严重不均衡时，F₁-score达到0.88。

关键词: 电池故障预警, 在线迁移学习, 下采样, 增量学习

Abstract:

Analyzing the characteristics of lithium batteries before failure is a complex task owing to their nature as electrochemical devices. Moreover, the limited number of failure samples in the production environment and the severe imbalance between positive and negative samples pose considerable challenges. To address these issues, this study proposes an improved online migration learning algorithm for early warning of high-voltage battery faults in batteries. First, to solve the problem of sample imbalance and reduce the use of computing resources, down-sampling technology is introduced. Specifically, a subsection down-sampling strategy is designed for the battery high-voltage fault early warning scenario. This strategy allows the algorithm model to learn more detailed features before the fault occurs. Second, an online transfer learning method based on batch incremental learning (homogeneous online transfer learning under incremental training, HomOTL-UIT) is proposed. This method addresses the need to update the offline classifier, which is trained in the source domain, at an appropriate time to adapt to the constantly changing data distribution in the target domain. Furthermore, it solves data distribution deviation issues and prevents the degeneration of online transfer learning into online learning. Batch incremental learning reduces the computing resource cost of multiple training, enabling continuous learning from the target domain and continuously improving the accuracy of offline classifiers over time. Then, a sliding window-based F₁-score scoring method is designed to solve the problem of the slow and unbalanced weight of the model to improve its accuracy. Finally, the validity and accuracy of the proposed method are verified using the operation data from an energy storage container. The results demonstrate its effectiveness, particularly when dealing with severely unbalanced positive and negative samples, thereby achieving an impressive F₁-score of 0.88.

Key words: battery failure warning, online transfer learning, lower sampling, incremental learning

中图分类号:

TM 912

董臣臣, 孙大帅, 王景龙. 基于改进的在线迁移学习算法的电池高压故障预警[J]. 储能科学与技术, 2023, 12(8): 2575-2584.

Chenchen DONG, Dashuai SUN, Jinglong WANG. Battery high-voltage fault early warning based on improved online migration learning algorithm[J]. Energy Storage Science and Technology, 2023, 12(8): 2575-2584.

图/表 13

图1

图2

图3

图4

表1

图5

表2

图6

表3

图7

图8

表4

表5

参考文献 23

1	FINEGAN D P, ZHU J E, FENG X N, et al. The application of data-driven methods and physics-based learning for improving battery safety[J]. Joule, 2021, 5(2): 316-329.
2	ZHU Y, LIU M Y, WANG L, et al. Potential failure prediction of lithium-ion battery energy storage system by isolation density method[J]. Sustainability, 2022, 14(12): 7048-7058.
3	LI D, ZHANG Z S, LIU P, et al. Battery fault diagnosis for electric vehicles based on voltage abnormality by combining the long short-term memory neural network and the equivalent circuit model[J]. IEEE Transactions on Power Electronics, 2021, 36(2): 1303-1315.
4	DUBARRY M, BECK D. Big data training data for artificial intelligence-based Li-ion diagnosis and prognosis[J]. Journal of Power Sources, 2020, 479: doi: 10.1016/j.jpowsour.2020.228806.
5	HU X S, ZHANG K, LIU K L, et al. Advanced fault diagnosis for lithium-ion battery systems: A review of fault mechanisms, fault features, and diagnosis procedures[J]. IEEE Industrial Electronics Magazine, 2020, 14(3): 65-91.
6	ZHAO J Y, LING H P, WANG J B, et al. Data-driven prediction of battery failure for electric vehicles[J]. iScience, 2022, 25(4): doi: 10.1016/j.isci.2022.104172.
7	王志福, 罗崴, 闫愿, 等. 基于GAPSO-FNN神经网络的锂离子电池传感器故障诊断[J]. 储能科学与技术, 2023, 12(2): 602-608.
	WANG Z F, LUO W, YAN Y, et al. Fault diagnosis of lithium ion battery sensor based on GAPSO-FNN neural network[J]. Energy Storage Science and Technology, 2023, 12(2): 602-608.
8	ZHANG Y Z, WIK T, BERGSTRÖM J, et al. A machine learning-based framework for online prediction of battery ageing trajectory and lifetime using histogram data[J]. Journal of Power Sources, 2022, 526: doi: 10.1016/j.jpowsour.2022.231110.
9	MAO W T, DING L, TIAN S Y, et al. Online detection for bearing incipient fault based on deep transfer learning[J]. Measurement, 2020, 152: doi: 10.1016/j.measurement.2019.107278.
10	RUAN H K, WEI Z B, SHANG W T, et al. Artificial Intelligence-based health diagnostic of lithium-ion battery leveraging transient stage of constant current and constant voltage charging[J]. Applied Energy, 2023, 336: doi: 10.1016/j.apenergy.2023.120751.
11	VON BÜLOW F, MEISEN T. State of health forecasting of heterogeneous lithium-ion battery types and operation enabled by transfer learning[J]. PHM Society European Conference, 2022, 7(1): 490-508.
12	CHE Y H, DENG Z W, LIN X K, et al. Predictive battery health management with transfer learning and online model correction[J]. IEEE Transactions on Vehicular Technology, 2021, 70(2): 1269-1277.
13	ODENA A. Semi-supervised learning with generative adversarial networks[EB/OL]. 2016: arXiv: 1606.01583. https: //arxiv.org/abs/1606.01583
14	HE J P, MAO R Y, SHAO Z M, et al. Incremental learning in online scenario[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020: 13923-13932.
15	VAN DE VEN G M, TUYTELAARS T, TOLIAS A S. Three types of incremental learning[J]. Nature Machine Intelligence, 2022, 4(12): 1185-1197.
16	ZHAO P L, HOI S C H, WANG J L, et al. Online transfer learning[J]. Artificial Intelligence, 2014, 216: 76-102.
17	WU Q Y, ZHOU X M, YAN Y G, et al. Online transfer learning by leveraging multiple source domains[J]. Knowledge and Information Systems, 2017, 52(3): 687-707.
18	HUANG S J, CAI N G, PACHECO P P, et al. Applications of support vector machine (SVM) learning in cancer genomics[J]. Cancer Genomics & Proteomics, 2018, 15(1): 41-51.
19	LIU K L, HU X S, ZHOU H Y, et al. Feature analyses and modeling of lithium-ion battery manufacturing based on random forest classification[J]. IEEE/ASME Transactions on Mechatronics, 2021, 26(6): 2944-2955.
20	CRAMMER K, DEKEL O, KESHET J, et al. Online passive aggressive algorithms[J]. Journal of Machine Learning Research, 2006, 7(1): 551-585.
21	JIANG J C, LI T Y, CHANG C, et al. Fault diagnosis method for lithium-ion batteries in electric vehicles based on isolated forest algorithm[J]. Journal of Energy Storage, 2022, 50: doi: 10.1016/j.est.2022.104177.
22	KRSTINIĆ D, BRAOVIĆ M, ŠERIĆ L, et al. Multi-label classifier performance evaluation with confusion matrix[C]// Computer Science & Information Technology. AIRCC Publishing Corporation, 2020: doi:10.5121/csit.2020.100801.
23	LI Y H, WANG N Y, SHI J P, et al. Adaptive batch normalization for practical domain adaptation[J]. Pattern Recognition, 2018, 80: 109-117.

符号	描述
X	源域
Y	目标域
v	离线分类器的预测函数
w	在线分类器的预测函数
Π	映射函数
B	在线分类器的批大小
τ	学习率
C	τ 的下限值
α₁、 α₂	分类器的权重系数
F_1,1	离线分类器的F₁-score
F_1,2	在线分类器的F₁-score

混淆矩阵		真实值
混淆矩阵		故障	正常
预测值	故障	TP	FP
预测值	正常	FN	TN

电芯	簇架	Pack	电芯数	温度传感器	电芯总数	高压阈值/V	电芯类型
电芯A	4	20	12	4	960	3.55	LiFePO₄
电芯B	2	16	15	4	480	3.55	LiFePO₄

混淆矩阵		真实值
混淆矩阵		故障	正常
预测值	故障	37	8
预测值	正常	2	26479
总计		39	26487

[1]	黄永浩, 臧国景, 朱霨亚, 廖友好, 李伟善. LiF添加剂改善含锂陶瓷隔膜与4.35 V LiNi_0.8Co_0.1Mn_0.1O₂ 正极的界面稳定性[J]. 储能科学与技术, 2023, 12(8): 2361-2369.
[2]	郭煜, 王亦伟, 钟隽, 杜进桥, 田杰, 李艳, 蒋方明. 基于增量容量曲线的锂离子电池微内短路故障诊断方法[J]. 储能科学与技术, 2023, 12(8): 2536-2546.
[3]	韩雨, 曹盛玲, 宁靖, 王康丽, 蒋凯, 周敏. 聚合物改性锂金属电池界面策略研究综述[J]. 储能科学与技术, 2023, 12(8): 2491-2503.
[4]	刘欢, 彭娜, 高清雯, 李文鹏, 杨志荣, 王景涛. 冠醚掺杂的聚合物固态电解质对全固态锂电池性能的影响[J]. 储能科学与技术, 2023, 12(8): 2401-2411.
[5]	左安昊, 方儒卿, 李哲. 锂离子电池单颗粒动力学表征方法综述[J]. 储能科学与技术, 2023, 12(8): 2457-2481.
[6]	李明, 谢金元, 邱沐楚, 邵亮, 霍强. 基于计划曲线的储能系统均衡热管理及节能研究[J]. 储能科学与技术, 2023, 12(8): 2585-2593.
[7]	狄云, 周正柱, 党会鸿, 葛志浩. 基于ECM的电芯电-热耦合建模与验证[J]. 储能科学与技术, 2023, 12(8): 2638-2648.
[8]	赵光金, 李博文, 胡玉霞, 董锐锋, 王放放. 退役动力电池梯次利用技术及工程应用概述[J]. 储能科学与技术, 2023, 12(7): 2319-2332.
[9]	范茂松, 耿萌萌, 赵光金, 杨凯, 王放放, 刘皓. 基于多频点阻抗的梯次利用电池分选技术研究[J]. 储能科学与技术, 2023, 12(7): 2202-2210.
[10]	宗磊, 陈龙, 朱峰, 刘国刚, 孙亚青, 董海书. 多场景下动力电池安全特征参数的阈值测试与分析[J]. 储能科学与技术, 2023, 12(7): 2271-2281.
[11]	李凌萱, 王子轩, 赵辰孜, 张睿, 卢洋, 黄佳琦, 陈爱兵, 张强. 复合金属锂负极的定量模型新进展[J]. 储能科学与技术, 2023, 12(7): 2059-2078.
[12]	陈智伟, 张维戈, 张珺玮, 张言茹. 基于视觉特征的动力电池组综合健康评估及分筛方法[J]. 储能科学与技术, 2023, 12(7): 2211-2219.
[13]	张宇波, 王有元, 黄洞宁, 王子懿, 陈伟根. 面向变工况条件的锂离子电池寿命退化预测方法[J]. 储能科学与技术, 2023, 12(7): 2238-2245.
[14]	武明虎, 岳程鹏, 张凡, 李俊晓, 黄伟, 胡胜, 唐靓. 多尺度分解下GRU-MLR组合的锂电池剩余使用寿命预测方法[J]. 储能科学与技术, 2023, 12(7): 2220-2228.
[15]	管鸿盛, 钱诚, 徐炳辉, 孙博, 任羿. 融合自注意力机制与门控循环单元网络的宽工况锂离子电池SOC估计[J]. 储能科学与技术, 2023, 12(7): 2229-2237.