基于集成学习与数据驱动的电动汽车动力电池多维度故障预警

doi:10.3969/j.issn.1674-8484.2024.03.010

摘要/Abstract

摘要：

针对汽车动力电池风险预测中单一算法在非均衡数据集上性能欠佳，以及分析维度不足等问题，提出一种建立在多维特征之上的数据驱动多模型集成学习方法以实现电池故障预警。根据内外部因素从电池状态、行驶状况、历史信息和时间环境多方面提取特征，以还原真实应用场景；通过Filter-Wrapper法剔除冗余无效信息来进行特征筛选，以提高鲁棒性；在Bayes优化结合网格搜索法完成超参数调优的基础上，构建了以SVM、LightGBM和XGBoost为初级学习器、LR为次级学习器的异质集成Stacking模型。结果表明：考虑了外部特征后的融合模型在评估指标上综合表现最好，其中在中、高等级的故障预警上召回率分别达到了85.81%和88.92%，相较传统仅考虑电池内部特征的单一模型有着更好的预测精度和泛化能力。

关键词: 汽车动力电池, 故障预警, Stacking集成学习, 多维度特征

Abstract:

Aiming at the problems such as poor performance of single algorithm on unbalanced data set and insufficient analysis dimension in electric vehicle power battery risk prediction, a data-driven multi-model ensemble learning method based on multi-dimensional characteristics was proposed to realize battery fault warning. According to internal and external factors, the features from battery information, driving condition, historical situation and time environment was extracted to simulate real application scenarios. To complete feature filtering, redundant information was removed by Filter-Wrapper method, which improved the robustness. The heterogeneous Stacking integrated model was constructed with SVM, LightGBM and XGBoost as primary learners and LR as secondary learners based on Bayesian optimization with grid search to optimize hyperparameters. The results show that the ensemble model with the addition of external features has the best comprehensive performance in the evaluation index, among which the recall rate in the medium and high level fault warning reaches 85.81% and 88.92%, respectively. Therefore, the method has better prediction accuracy and generalization ability than the single model that only considers the internal characteristics of the battery.

Key words: vehicle power battery, early warning of fault, Stacking ensemble learning, multi-dimensional feature

中图分类号:

U472
TM911

王健俊, 陈豪, 付元承. 基于集成学习与数据驱动的电动汽车动力电池多维度故障预警[J]. 汽车安全与节能学报, 2024, 15(3): 368-378.

WANG Jianjun, CHEN Hao, FU Yuancheng. Data-driven fault early warning of electric vehicle power battery based on ensemble learning from multiple dimensions[J]. Journal of Automotive Safety and Energy, 2024, 15(3): 368-378.

图/表 17

参考文献 16

[1]	WANG Zhenpo, HONG Jichao, LIU Peng, et al. Voltage fault diagnosis and prognosis of battery systems based on entropy and Z-score for electric vehicles[J]. Appl Energ, 2017, 196: 289-302.
[2]	HONG Jichao, WANG Zhenpo, LIU Peng. Big-data-based thermal runaway prognosis of battery systems for electric vehicles[J]. Energies, 2017, 10(7): 919.
[3]	WANG Yujie, TIAN Jiaqiang, CHEN Zonghai, et al. Model based insulation fault diagnosis for lithium-ion battery pack in electric vehicles[J]. Measurement, 2019, 131: 443-451. doi: 10.1016/j.measurement.2018.09.007
[4]	刘鹏, 朱建新, 储爱华, 等. 基于热模型的动力电池热故障诊断系统[J]. 上海交通大学学报, 2015, 49(4): 487-493.
	LIU Peng, ZHU Jianxin, CHU Aihua, et al. Power battery thermal faults diagnostic system based on thermal model[J]. J Shanghai Jiaotong Univ, 2015, 49(4): 487-493. (in Chinese)
[5]	LI Da, ZHANG Zhaosheng, LIU Peng, 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 Trans Power Electr, 2021, 36(2): 1303-1315.
[6]	WANG Jia, ZHANG Shenglong, HU Xia. A fault diagnosis method for lithium-ion battery packs using improved RBF neural network[J]. Front Energ Res, 2021, 9: 702139.
[7]	YAO Lei, FANG Zhangpeng, XIAO Yanqiu, et al. An intelligent fault diagnosis method for lithium battery systems based on grid search support vector machine[J]. Energy, 2021, 214: 118866.
[8]	周仁, 张向文. 基于实车数据和BP-AdaBoost算法的电动汽车动力电池健康状态估计[J]. 科学技术与工程, 2022, 22(21): 9398-9406.
	ZHOU Ren, ZHANG Xiangwen. Electric vehicle power battery SOH estimation based on real vehicle data and BP-Adaboost algorithm[J]. Sci Tech Engi, 2022, 22(21): 9398-9406. (in Chinese)
[9]	胡杰, 朱雪玲, 何陈, 等. 基于实车数据的电动汽车电池健康状态预测[J]. 汽车工程, 2021, 43(9): 1291-1299.
	HU Jie, ZHU Xueling, HE Chen, et al. Prediction on battery state of health of electric vehicles based on real vehicle data[J]. Autom Engineering, 2021, 43(9): 1291-1299. (in Chinese)
[10]	罗勇, 祁朋伟, 黄欢, 等. 基于容量修正的安时积分SOC估算方法研究[J]. 汽车工程, 2020, 42(5): 681-687.
	LUO Yong, QI Pengwei, HUANG Huan, et al. Study on battery SOC estimation by ampere-hour integral method with capacity correction[J]. Autom Engineering, 2020, 42(5): 681-687. (in Chinese)
[11]	颜湘武, 郭琪, 杨漾, 等. 动力电池组健康状态评价方法的研究[J]. 湖南大学学报(自然科学版), 2015, 42(2): 93-99.
	YAN Xiangwu, GUO Qi, YANG Yang, et al. Study on methods for estimating the state of health of battery pack[J]. J Hunan Univ (Nat Sci Edit), 2015, 42(2): 93-99. (in Chinese)
[12]	金思含, 彭忆强, 武小花, 等. 成都市燃料电池公交车示范运行行驶工况的构建[J]. 汽车安全与节能学报, 2022, 13(1): 202-208.
	JIN Sihan, PENG Yiqiang, WU Xiaohua, et al. Construction of the driving cycle for fuel cell bus running in Chengdu demonstration area[J]. J Autom Safe Energ, 2022, 13(1): 202-208. (in Chinese)
[13]	李明俊, 张正豪, 宋晓琳, 等. 基于一种多分类半监督学习算法的驾驶风格分类模型[J]. 湖南大学学报(自然科学版), 2020, 47(4): 10-15.
	LI Mingjun, ZHANG Zhenghao, SONG Xiaolin, et al. Driving style classification model based on a multi-label semi-supervised learning algorithm[J]. J Hunan Univ (Nat Sci Edit), 2020, 47(4): 10-15. (in Chinese)
[14]	Nti I K, Adekoya A F, Weyori B A. A comprehensive evaluation of ensemble learning for stock-market prediction[J]. J Big Data, 2020, 7(1): 1-40.
[15]	Krogh A, Vedelsby J. Neural network ensembles, cross validation, and active learning[J/OL]. Neur Info Proc Syst, 1994, 7. [2023-07-04]. https://proceedings.neurips.cc/paper/1994/hash/b8c37e33defde51cf91e1e03e51657da-Abstract.html.
[16]	Ting K M, Witten I H. Issues in stacked generalization[J]. J Artif Intell Res, 1999, 10: 271-289.

数据类别	数据字段名称	字段定义	格式备注
车辆信息	vid	车辆ID	XX xxx
外部环境信息	yr_modahrmn	数据采集时间	年-月-日- 时:分:秒
车辆运行状态	charging_status	充电状态	01停车充电； 02行驶充电； 03未充电； 04充电完成
动力电池信息	standard_soc	SOC—国标	0~100%，变动单位1%
车辆故障情况	alarm_info	通用报警标志	10进制转化为2进制后查表

数据类别	数据字段名称	字段定义	格式备注
车辆信息	vid	车辆ID	XX xxx
外部环境信息	yr_modahrmn	数据采集时间	年-月-日- 时:分:秒
车辆运行状态	charging_status	充电状态	01停车充电； 02行驶充电； 03未充电； 04充电完成
动力电池信息	standard_soc	SOC—国标	0~100%，变动单位1%
车辆故障情况	alarm_info	通用报警标志	10进制转化为2进制后查表

标志位	故障等级	定义	报警标准(动态)	报警依据
0	1	1：电池低温报警；0：正常	<15 ℃	容量衰减，内阻增大，产热率增加
1	1	1：电池高温报警；0：正常	>50 ℃	加速老化，Arrhenius定律
2	1	1：温度差异报警；0：正常	>5 ℃	电池内部温度差异范围
3	3	1：电池极限低温报警；0：正常	<5 ℃	负极表面折锂，液态电解质或结晶
4	3	1：电池极限高温报警；0：正常	>100 ℃	电池吉速老化，负极表面SEI分解
5	3	1：温度极限差异报警；0：正常	>20 ℃	电池内部温度差异范围
6	2	1：单体电池欠压报警；0：正常	<3 V	NMC电池产品信息
7	3	1：单体电池过压报警；0：正常	>4.3 V	NMC电池产品信息
8	3	1：单体电池过充；0：正常	充电过程中 >4.25 V	NMC电池产品信息
9	2	1：单体电池一致性差报警；0：正常	单体电压标准差>20 mV	电池组内所有单体工作电压标准差
10	1	1：SOC低报警；0：正常	<10%	经验值
11	1	1：SOC跳变报警；0：正常	ΔSOC>5%	经验值
12	1	1：绝缘报警；0：正常	直流绝缘电阻<100 Ω/V	国标
13	2	1：DC-DC状态报警；0：正常	DC-DC状态断开且未熄火	经验值
14		预留

标志位	故障等级	定义	报警标准(动态)	报警依据
0	1	1：电池低温报警；0：正常	<15 ℃	容量衰减，内阻增大，产热率增加
1	1	1：电池高温报警；0：正常	>50 ℃	加速老化，Arrhenius定律
2	1	1：温度差异报警；0：正常	>5 ℃	电池内部温度差异范围
3	3	1：电池极限低温报警；0：正常	<5 ℃	负极表面折锂，液态电解质或结晶
4	3	1：电池极限高温报警；0：正常	>100 ℃	电池吉速老化，负极表面SEI分解
5	3	1：温度极限差异报警；0：正常	>20 ℃	电池内部温度差异范围
6	2	1：单体电池欠压报警；0：正常	<3 V	NMC电池产品信息
7	3	1：单体电池过压报警；0：正常	>4.3 V	NMC电池产品信息
8	3	1：单体电池过充；0：正常	充电过程中 >4.25 V	NMC电池产品信息
9	2	1：单体电池一致性差报警；0：正常	单体电压标准差>20 mV	电池组内所有单体工作电压标准差
10	1	1：SOC低报警；0：正常	<10%	经验值
11	1	1：SOC跳变报警；0：正常	ΔSOC>5%	经验值
12	1	1：绝缘报警；0：正常	直流绝缘电阻<100 Ω/V	国标
13	2	1：DC-DC状态报警；0：正常	DC-DC状态断开且未熄火	经验值
14		预留

类别	聚类中心	工况特点
0	(22.38 78.3 0.214)	较少停驻、车速中等
1	(52.32 92.4 0.251)	几乎一直通行、车速较高
2	(15.62 39.3 0.120).	停驻时间占比大、车速较低