Research on SOC estimation of lithium-ion power battery based on feature combination and stacking fusion ensemble Learning

doi:10.19799/j.cnki.2095-4239.2020.0132

Abstract

Abstract:

Studies on the estimation of the battery power state of charge (SOC) are primarily based on the charge and discharge data of a single cell under ideal experimental conditions, which may not correspond to the actual complex and variable driving conditions. In response to this problem, relying on the National Big Data Alliance of New Energy Vehicles and using data-driven methods, the transient, dynamic, and driving behavior and aging, dynamic, time-varying, and mileage features were derived. Using the method of stacking fusion and integration learning based on feature combination, an SOC estimation of the battery power discharge process under actual complex and variable working conditions was conducted. A reference XGBoost Model 0 based on transient features, five XGBoost Models (1, 2, 3, 5, and 6) based on feature combination, and a Linear Model 4 were constructed for comparison and analysis. Comparing Models 1-6 with the reference Model 0, the mean absolute error (MAE) and mean square error (MSE) of the stacked fusion model (stacked model) based on feature combination were the smallest, 0.39 and 0.32, respectively, which compared to Model 0, decreased by 38% and 46%, respectively; meanwhile, the coefficient of determination was the highest, reaching 0.9995, which was an increase of 2% compared to Model 0. The generalization ability of the stacked model also performed well, and the average and standard deviation of its accuracy reached 98.89% and 0.03%, respectively. Comparing Models 5 and 6 and the reference Model 0, it can be seen that, as the feature dimension increases, the MAE and MSE of the model decreases, the coefficient of determination increases, and the performance of the model improves. This study helps promote the application of data-driven methods in the estimation of the SOC of power batteries and provides guidance and a reference for the estimation of the SOC of actual electric vehicles.

Key words: power battery, Xgboost, SOC estimation, ensemble learning, feature combination

CLC Number:

TK 9

Ying HE, Genpeng ZHONG, Yi CHEN. Research on SOC estimation of lithium-ion power battery based on feature combination and stacking fusion ensemble Learning[J]. Energy Storage Science and Technology, 2020, 9(5): 1548-1557.

Figures/Tables 18

Fig.1

Table 1

Typical data record fragments"

属性	范例	说明
time	2018-07-11 12:05:40	2018年7月11日12点5分40秒
SOC	78	剩余电量占总电量百分比
CS	1	1-充电，3-行驶，4-充满
V	351.3	电池总电压/V
L	9457	行驶里程/km
I	20	母线电流/A
$V c m a x$	4.021	单体最高电压/V
$V c m i n$	3.998	单体最低电压/V
$T c m a x$	28	单体最高温度/℃
$T c m i n$	27	单体最低温度/℃

Table 1

Table 2

Normal data reference range"

属性	范围
SOC	0～100
V/V	300～380
I/A	-200～300
$V c m a x$ ， $V c m i n$ /V	3.3～4.2
$T c m a x$ ， $T c m i n$ /℃	-15～50

Table 2

Fig.2

Fig.3

Table 3

Table 4

Table 5

Fig.4

Table 6

Fig.5

Fig.6

Fig.7

Fig.8

Fig.9

Table 7

The training result of the stacking model"

模型	特征集合	MAE	MSE	$R 2$
Xgboost Model 0	瞬态特征	0.63	0.59	0.9788
Xgboost Model 1	瞬态+驾驶行为特征	0.47	0.46	0.9913
Xgboost Model 2	瞬态+衰老特征	0.61	0.56	0.9802
Xgboost Model 3	瞬态+时变特征	0.45	0.40	0.9843
Linear Model 4	瞬态+行驶里程特征	0.55	0.51	0.9842
Xgboost Model 5	瞬态+动态特征	0.44	0.39	0.9914
Xgboost Model 6	全部特征	0.42	0.34	0.9949
Xgboost Model 7	全部特征+全部预测值	0.39	0.32	0.9995

Table 7

Fig.10

Table 8

References 18

1	新能源汽车国家大数据联盟. 中国新能源汽车大数据研究报告[M]. 北京：社会科学文献出版社, 2019: 1-10.
2	李靖. 基于UKF算法的电动汽车动力电池荷电状态的研究[D]. 西安: 西安理工大学, 2013.
3	周美兰, 赵强, 周永琴. 改进的PSO-BP神经网络估算磷酸铁锂电池SOC[J]. 哈尔滨理工大学学报, 2015, 20(4): 88-92.
	ZHOU Meilan, ZHAO Qiang, ZHOU Yongqin. State of charge estimation for LiFePO₄ cell using improved PSO-BP neural network[J]. Journal of Harbin University of Science and Technology, 2015, 20(4): 88-92.
4	GUO Y, ZHAO Z, LUO M. SOC estimation of lithium battery based on improved BP neural network[J]. Energy Procedia, 2017, 105: 4153-4158.
5	孙豪豪, 潘庭龙, 吴定会. 基于自适应电池模型的SOC加权在线估计[J]. 系统仿真学报, 2017, 8(29): 1677-1675.
	SUN Haohao, PAN Tinglong, WU Dinghui. Weighted online estimation of SOC based on adaptive battery model[J]. Journal of System Simulation, 2017, 8(29): 1677-1675.
6	孙志祥. 考虑温度影响的动力锂电池SOC估计算法和验证[D]. 长春: 吉林大学, 2018.
7	康燕琼. 纯电动汽车锂电池组健康状态（SOH）的估计研究[D]. 北京: 北京交通大学, 2015.
	KANG Yanqiong. Research of estimation for SOH of PEV Li-ion battery pack[J]. Beijing: Beijing Jiaotong University, 2015.
8	CHEN Tianqi, GUESTRIN Carlos. Xgboost: A scalable tree boosting system[C]//ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016.
9	WOLPERT D H. Stacked generalization[M]. USA: Springer, 2011.
10	刘毅, 谭国俊, 何晓群. 优化电池模型的自适应Sigma卡尔曼荷电状态估算[J]. 电工技术学报, 2017, 20(4): 108-118.
	LIU Yi, TAN Guojun, HE Xiaoqun. Optimized battery model based adaptive sigma Kalman filter for state of charge estimation[J]. Transactions of China Electrotechnical Society, 2017, 20(4): 108-118.
11	廖恩华. 基于神经网络的电动汽车磷酸铁锂电池SOC估算方法研究[D]. 成都: 电子科技大学, 2011.
12	郝中华. 基于主元分析的锂动力电池SOC估算方法研究[D]. 哈尔滨: 哈尔滨理工大学, 2014.
	HAO Zhonghua. Lithium battery SOC estimation method study based on principal component anaylsis[D]. Harbin: Harbin University of Science and Technology, 2014.
13	杨辉宝. 福州市区典型道路汽车行驶工况的研究[D]. 福州: 福州大学, 2017.
14	田宇. 太原市轻型车道路行驶工况构建与分析[D]. 太原: 太原理工大学, 2018.
15	刘存山. 基于NEDC工况的动力电池使用寿命测试方法研究[J]. 电源技术, 2017, 41(11): 47-49.
	LIU Cunshan. Research on battery life test method based on NEDC cycle[J]. Chinese Journal of Power Sources, 2017, 41(11): 47-49.
16	张憧. 电动汽车续驶里程影响因素及预测研究[D]. 合肥: 合肥工业大学, 2018.
17	谢明维. 纯电动汽车剩余续驶里程估算研究[D]. 镇江: 江苏大学, 2016.
	XIE Weiming. Research on remaining driving range estimation of battery electric vehicle[D]. Zhenjiang: Jiangsu University, 2016.
18	李艳芳, 王钰, 李济洪. 几种交叉验证检验的可重复性[J]. 太原师范学院学报(自然科学版), 2013, 12(4): 46-49.
	LI Yanfang, WANG Yu, LI Jihong. The replicability of several cross-validated tests[J]. Journal of Teiyuan Normal University(Natural Science Edition), 2013, 12(4): 46-49.

参数	说明	参数	说明
T_c	匀速时间/s	a_min	最小加速度/m·s^-2
T_d	减速时间/s	a_max	最大加速/m·s^-2
T_a	加速时间/s	a_a	加速段平均加速度/m·s^-2
T	总时间/s	a_d	减速段平均减速度/m·s^-2
L	运行距离/m	a_msd	加速度标准差/m·s^-2
v_mavg	行驶平均车速/km·h^-1	I_max	最大电流/A
v_max	最大车速/km·h^-1	I_m	电流标准差/A
v_msd	行驶速度标准差/km·h^-1	I_sd	平均电流/A
v_sd	速度标准差/km·h^-1

参数	说明	参数	说明
P_a	加速时间	P_{60_70}	60～70 km/h车速比例
P_d	减速时间	P_＞70	＞70 km/h车速比例
P_c	匀速时间	A₃₂	减速段减速度在-3～-2 m/s²比例
P_{0_10}	0～10 km/h车速比例	A₂₁	减速段减速度在-2～-1 m/s²比例
P_{10_20}	10～20 km/h车速比例	A_{1_01}	减速段减速度在-1～0.1 m/s²比例
P_{20_30}	20～30 km/h车速比例	A_{01_1}	加速段减速度在0.1～1 m/s²比例
P_{30_40}	30～40 km/h车速比例	A_{1_2}	加速段减速度在1～2 m/s²比例
P_{40_50}	40～50 km/h车速比例	A_{2_3}	加速段减速度在2～3 m/s²比例
P_{50_60}	50～60 km/h车速比例

参数	1	2	3	4	5	6
T_c	45.4	42.6	31.8	25.5	245.0	33.6
T_d	28.9	50.1	35.7	48.4	216.4	49.7
T_a	29.8	51.1	52.2	44.0	236.6	69.3
T	104.1	143.8	119.7	117.9	698.0	152.6
L	269.9	830.3	873.7	944.4	9523.8	2173.1
v_mavg	10.3	21.7	27.2	30.5	49.7	49.5
v_max	22.8	41.5	52.1	59.1	84.5	88.1
v_msd	7.1	12.6	16.8	18.6	20.6	25.6
v_sd	7.6	13.9	18.4	20.5	21.8	29.3
a_min	-0.47	-0.75	-1.10	-1.14	-1.09	-2.21
a_max	0.45	0.72	0.72	1.23	0.94	1.65
a_a	0.27	0.37	0.37	0.57	0.31	0.63
a_d	-0.28	-0.37	-0.56	-0.53	-0.34	-0.86
a_msd	0.05	0.05	0.05	0.06	0.06	0.06
I_max	21.3	46.8	54.0	80.2	110.0	126.1
I_m	4.9	9.2	11.2	13.4	19.7	27.9
I_sd	7.7	17.3	22.5	32.1	33.2	48.7
P_a	28.7%	35.4%	44.6%	37.4%	34.5%	47.7%
P_d	28.0%	35.0%	29.7%	41.4%	31.7%	35.0%
P_c	43.3%	29.6%	25.7%	21.3%	33.7%	17.3%
P_{0_10}	61.5%	34.1%	36.5%	35.5%	9.7%	25.2%
P_{10_20}	28.2%	16.3%	11.3%	11.0%	8.4%	7.7%
P_{20_30}	10.0%	23.5%	11.6%	10.3%	8.7%	5.9%
P_{30_40}	1.2%	19.1%	16.0%	11.2%	8.5%	5.8%
P_{40_50}	0.1%	6.3%	15.2%	14.9%	11.7%	6.4%
P_{50_60}	0.0%	1.0%	7.3%	10.9%	15.7%	10.7%
P_{60_70}	0.0%	0.3%	2.1%	4.7%	15.8%	13.9%
P_＞70	0.0%	0.1%	1.0%	2.4%	21.8%	25.3%
A₃₂	0.0%	0.0%	0.0%	0.0%	0.0%	3.8%
A₂₁	0.0%	0.4%	4.0%	5.1%	0.9%	8.6%
A_{1_01}	72.0%	64.8%	52.0%	58.2%	64.7%	40.4%
A_{01_1}	29.1%	35.5%	44.8%	31.3%	34.1%	38.7%
A_{1_2}	0.0%	0.2%	0.2%	6.4%	0.5%	8.5%
A_{2_3}	0.0%	0.0%	0.0%	0.0%	0.0%	1.0%

参数	说明	取值范围	本文取值
学习率Eta	控制学习速度，收缩更新步长	[0,1]	0.02
最小孩子权重Min_child_weight	防止过拟合	[0, ∞]	1
最大树深度 Max_depth	防止过拟合	[1, ∞]	6
子采样 Subsample	低值使得模型更保守且能防止过拟合	[0.5,1]	0.6
列采样 Colsample_bytree	每棵树随机选取的特征比例	[0.5,1]	0.8
L₁正则参数 alpha	当数据维度较高时，使得算法运行更快	[0, ∞]	1
L₂正则参数 Lambda	防止过拟合	[0, ∞]	2

模型	准确性平均值/%	准确性标准差/%
Model 0	96.02	0.18
Model 1	95.68	0.21
Model 2	96.13	0.19
Model 3	95.57	0.12
Model 5	97.43	0.11
Model 6	95.98	0.08
Model 7	98.89	0.03