基于BP神经网络的电动汽车动力电池产热估计

doi:10.16085/j.issn.1000-6613.2023-0229

摘要/Abstract

摘要：

电池的产热情况是电池热管理的重要指标之一，准确估计电池产热功率对构建高效运行的电池热管理系统以确保电动汽车安全行驶至关重要。然而，目前大多采用基于模型的方法进行电池产热估计，但此方法存在耗费时间长和仅应用于某种特定电池状况产热估计等缺点，无法解决电动汽车电池实时产热估计的问题。对此，本文提出了一种基于人工智能算法的精准电池产热功率估计方法，即基于BP（back propagation，BP）神经网络的电动汽车动力电池产热估计模型。该模型利用贝叶斯优化算法（Bayesian optimization，BO）对BP神经网络进行超参数选取，采用Adam（adaptive momentum estimation，Adam）优化算法加快收敛速度，提高了模型的准确度和稳定性。研究对比了不同放电倍率和不同环境温度下恒流放电实验的电池产热功率，结果表明模型的估计平均误差为5.01%，最大误差仅为5.53W，R²拟合指标最高可达99.98%，证明了所提出的电池产热估计模型取得了较高的估计精度和较强的鲁棒性，为电动汽车电池实时产热估计提供了一个范式结构。

关键词: 电动汽车, 动力电池, BP神经网络, 产热估计, 优化算法

Abstract:

The heat generation of battery is one of the critical indicators of battery thermal management. Accurate estimation of the heat generation rate of the battery is crucial to building an efficient battery thermal management system and thereby facilitating the safe driving of electric vehicles. However, most researchers currently rely on model-based simulations to estimate the heat generation rate of electric vehicle batteries. There are some disadvantages in this method, such as time-consuming and only application to some specific battery condition, which impede its wide application in addressing real-time heat generation rate of battery for electric vehicles. This paper proposed a precise battery heat production rate estimation strategy based on BO-Adam-BP neural network approach, that was, an electric vehicle power battery heat production estimation model based on BP neural network. The model used the Bayesian optimization algorithm (Bayesian optimization, BO) to select hyperparameters of BP (back propagation, BP) neural network, and used Adam (adaptive momentum estimation) optimization algorithm to speed up the convergence speed and improve the accuracy and stability of the model. Comparing to the battery heat generation power of constant current discharge experiments under different discharge rates and various ambient temperatures, the results showed that the estimated average error of the model was 5.01%, and the maximum error was only 5.53W. The R² fitting index could reach up to 99.98%, proving that the proposed battery heat production estimation model had achieved high accuracy and strong robustness, providing a paradigm structure for real-time heat production estimation of electric vehicle batteries.

Key words: electric vehicle, power battery, BP neural network, heat generation rate estimation, optimization algorithm

中图分类号:

TK11⁺3

王敬翰, 吕杰, 赵丁, 林文野, 宋文吉, 冯自平. 基于BP神经网络的电动汽车动力电池产热估计[J]. 化工进展, 2024, 43(1): 400-406.

WANG Jinghan, LYU Jie, ZHAO Ding, LIN Wenye, SONG Wenji, FENG Ziping. BP neural network approach for heat generation rate estimation of power battery for electric vehicles[J]. Chemical Industry and Engineering Progress, 2024, 43(1): 400-406.

图/表 13

表1 电池产热估计方法优缺点概述

方法	优点	缺点	准确度	稳定性
基于实验的方法	实验系统简单，人为操作不复杂	需要耗费大量的时间，仅限于实验室条件	★★★★	★★★
基于模型的方法	产热机理解释清晰，能了解内在逻辑关系	仿真结果只表示某种特定条件，不适用复杂行驶条件	★★★	★★★★
基于人工智能算法的方法	模型具有通用性和可迁移性^[5]，适用复杂行驶条件	需要大量数据支撑，系统原理复杂	★★★★★	★★★★

图1 基于BP神经网络的锂电池产热估计模型

图2 基于BP神经网络的锂电池产热估计模型流程

表2 贝叶斯优化算法流程[13]

序号	流程
1	For t=1, 2, …, do
2	最小化采集函数，得到下一个评估点：x_t =argmin _x_∈_Uz(x\|D_1:_t_-1)
3	评估目标函数值：y_t =f(x_t )+ε_t
4	整理数据：D_1:_t =D_1:_t_-1∪(x_t, y_t )
5	更新目标函数f(x)
6	End for

表3 超参数寻优选取范围

参数	数值范围
批量处理样本数	[1000,1500]
L2正则化系数	[10^-8, 10^-3]
初始学习率	[10^-3, 10^-1]

表4 Adam优化算法流程

序号	流程
1	While θ_t 不收敛 do
2	更新步长：t=t+1
2	计算损失函数梯度：g_t =∇ _wf(θ_t_-1)
3	计算一阶矩和二阶矩：m_t =β₁m_t_-1+(1-β₁)g_t，v_t =β₂m_t_-1+(1-β₂)g_t²
4	偏差纠正： $z m t$ = $z m t$ /(1-β₁^t )， $z v t$ = $z v t$ /(1-β₂^t )
5	更新权值w_t 和阈值b_t ： $θ t = θ t - 1 - α z m t z v t + ε$
6	End while

表4 Adam优化算法流程

序号	流程
1	While θ_t 不收敛 do
2	更新步长：t=t+1
2	计算损失函数梯度：g_t =∇ _wf(θ_t_-1)
3	计算一阶矩和二阶矩：m_t =β₁m_t_-1+(1-β₁)g_t，v_t =β₂m_t_-1+(1-β₂)g_t²
4	偏差纠正： $z m t$ = $z m t$ /(1-β₁^t )， $z v t$ = $z v t$ /(1-β₂^t )
5	更新权值w_t 和阈值b_t ： $θ t = θ t - 1 - α z m t z v t + ε$
6	End while

表5 三元锂电池技术参数

项目	参数
额定容量/A·h	40
额定电压/V	3.7
放电能量/W·h	143
充电截止电压/V	4.2
放电截止电压/V	2.75
长×宽×高/mm	148×27×92
电池质量/g	840
比热容/J·kg^-1·K^-1	1192.16
热导率/W·m^-1·K^-1	沿长度方向：55.5
	沿宽度方向：0.8531
	沿高度方向：55.5

图3 电池产热测试实验系统

表6 部分电池参数数据

数据编码	U_oc/V	U_act/V	T_bat/℃	Q/W
1	4.1164	4.0008	25.687	4.6244
2	4.1161	4.0004	25.71	4.6281
3	4.1158	4.0001	25.702	4.6278
4	4.1155	3.9998	25.694	4.6275
5	4.1152	3.9995	25.71	4.6273
6	4.1149	3.9992	25.699	4.6270
7	4.1146	3.9989	25.71	4.6268
8	4.1143	3.9986	25.72	4.6267
9	4.1140	3.9983	25.728	4.6265
10	4.1137	3.9977	25.715	4.6384

表7 PCA主成分分析结果

主成分	指标	累计贡献率
x₁	U_act/V	0.9580
x₂	I/A	0.9862
x₃	SOC	1
x₄	电池表面温度/℃	1

图4 恒流放电过程产热量预测值对比

表8 不同放电倍率预测误差表

倍率	MADE	RMSE	R²
0.5C	0.0852	0.1191	0.9945
1.0C	0.0511	0.0637	0.9998
1.5C	0.1963	0.3813	0.9948
2.5C	0.3900	0.5831	0.9961

表9 不同温度预测误差表

温度/℃	MADE	RMSE	R²
0	0.0723	0.1246	0.9983
25	0.0852	0.1191	0.9998
35	0.1235	0.1849	0.9862

参考文献 16

1	HU Yang, CHOE Song-Yul, GARRICK T R. Measurement of two-dimensional heat generation rate of pouch type lithium-ion battery using a multifunctional calorimeter[J]. Journal of Power Sources, 2022, 532: 231350.
2	HU Yang, CHOE Song-Yul. Simultaneous and continuous characterization of reversible and irreversible heat of lithium-ion battery using wavelet transform technique[J]. Electrochimica Acta, 2021, 375: 137973.
3	XU Xiaobin, ZHANG Hengyun, LIU Shunbo, et al. Surrogate models for lithium-ion battery heat generation based on orthogonal experiments by eliminating external wire connection effect[J]. Applied Thermal Engineering, 2022, 213: 118655.
4	REN Honglei, JIA Li, DANG Chao, et al. An electrochemical-thermal coupling model for heat generation analysis of prismatic lithium battery[J]. Journal of Energy Storage, 2022, 50: 104277.
5	ZHU Shan, HE Chunnian, ZHAO Naiqin, et al. Data-driven analysis on thermal effects and temperature changes of lithium-ion battery[J]. Journal of Power Sources, 2021, 482: 228983.
6	ARORA S, SHEN Weixiang, KAPOOR A. Neural network based computational model for estimation of heat generation in LiFePO₄ pouch cells of different nominal capacities[J]. Computers & Chemical Engineering, 2017, 101: 81-94.
7	PANG Hui, WU Longxing, LIU Jiahao, et al. Physics-informed neural network approach for heat generation rate estimation of lithium-ion battery under various driving conditions[J]. Journal of Energy Chemistry, 2023, 78: 1-12.
8	YALÇIN S, PANCHAL S, HERDEM M S. A CNN-ABC model for estimation and optimization of heat generation rate and voltage distributions of lithium-ion batteries for electric vehicles[J]. International Journal of Heat and Mass Transfer, 2022, 199: 123486.
9	MANOHARAN A, BEGAM K K M, APAROW V R, et al. Artificial Neural Networks, Gradient Boosting and Support Vector Machines for electric vehicle battery state estimation: A review[J]. Journal of Energy Storage, 2022, 55: 105384.
10	BERNARDI D, PAWLIKOWSKI E, NEWMAN J. A general energy balance for battery systems[J]. Journal of the Electrochemical Society, 1985, 132(1): 5-12.
11	HE Feifei, ZHOU Jianzhong, FENG Zhongkai, et al. A hybrid short-term load forecasting model based on variational mode decomposition and long short-term memory networks considering relevant factors with Bayesian optimization algorithm[J]. Applied Energy, 2019, 237: 103-116.
12	PELIKAN M, GOLDBERG D E, CANTÚ-PAZ E. BOA: The Bayesian optimization algorithm[C]// Proceedings of the Gene-tic and Evolutionary Computation Conference GECCO-99. 1999: 525-532.
13	李亚茹, 张宇来, 王佳晨. 面向超参数估计的贝叶斯优化方法综述[J]. 计算机科学, 2022, 49(S1): 86-92.
	LI Yaru, ZHANG Yulai, WANG Jiachen. Survey on Bayesian optimization methods for hyper-parameter tuning[J]. Computer Science, 2022, 49(S1): 86-92.
14	KINGMA D P, BA J. Adam: A method for stochastic optimization[C]// International Conference on Learning Representations 2015, San Diego, 2015.
15	李扬. 动力锂离子电池电学参数及热物性测量方法研究[D]. 北京: 中国科学院大学, 2019.
	LI Yang. Study on measurement methods of electrical parameters and thermophysical properties of power lithium-ion batteries[D]. Beijing: University of Chinese Academy of Sciences, 2019.
16	李晟延, 马鸿雁, 窦嘉铭, 等. 基于PBES-LS-SVM的锂离子电池组SOC预测[J]. 电源技术, 2022, 46(11): 1279-1283.
	LI Shengyan, MA Hongyan, DOU Jiaming, et al. SOC prediction of Li-ion battery pack based on PBES-LS-SVM[J]. Chinese Journal of Power Sources, 2022, 46(11): 1279-1283.

[1]	潘艳秋, 李鹏飞, 高石磊, 俞路. 基于MIC筛选规则和BP神经网络的变换装置建模及产品预测[J]. 化工进展, 2022, 41(S1): 36-43.
[2]	贾文龙, 孙溢彬, 汤丁, 陈家文, 雷思罗, 李长俊. 基于支持向量机的输气管道泄漏压降信号智能识别方法[J]. 化工进展, 2022, 41(9): 4713-4722.
[3]	罗明昀, 凌子夜, 方晓明, 张正国. 基于相变储热技术的电池热管理系统研究进展[J]. 化工进展, 2022, 41(3): 1594-1607.
[4]	苏戈曼, 崔国民, 肖媛, 赵倩倩. 换热网络优化中换热单元生成频次的影响分析及策略改进[J]. 化工进展, 2020, 39(10): 3879-3891.
[5]	韩丰磊, 李梦雨, 刘杰夫, 李丹丹, 郭雯雯, 周硕, 赵朝成, 赵东风. 基于BP神经网络的煤直接液化装置VOCs包袋法检测与预测建模[J]. 化工进展, 2020, 39(10): 3949-3955.
[6]	杨泛明,焦奇方,伍伟,贺国文. 涂碳铝箔对LiFePO₄动力电池性能影响[J]. 化工进展, 2019, 38(10): 4639-4644.
[7]	贺瑞军, 邹得球, 马先锋, 刘小诗, 郭江荣, 胡志钢, 刘默. 基于碳纳米材料协同强化型复合相变材料的动力电池组传热特性[J]. 化工进展, 2018, 37(11): 4174-4180.
[8]	宋俊杰, 王义春, 王腾. 动力电池组分层风冷式热管理系统仿真[J]. 化工进展, 2017, 36(S1): 187-194.
[9]	华政, 梁风, 姚耀春. 电动汽车电池的发展现状与趋势[J]. 化工进展, 2017, 36(08): 2874-2881.
[10]	付柯, 谢良才, 闫雨瑗, 李波, 贺改, 徐龙, 马晓迅. 改进BP神经网络预测Ni/Al₂O₃催化CH₄-CO₂重整反应[J]. 化工进展, 2017, 36(07): 2393-2399.
[11]	李振宇, 任文坡, 黄格省, 金羽豪, 师晓玉. 我国新能源汽车产业发展现状及思考[J]. 化工进展, 2017, 36(07): 2337-2343.
[12]	靳鹏超1，王世学1，2. 一种使用相变材料的新型电动汽车电池热管理系统 [J]. 化工进展, 2014, 33(10): 2608-2612.
[13]	赵斌1,3，吴桂芳2，郝培文1. 应用红外光谱进行不同性状沥青的判别[J]. 化工进展, 2012, 31(12): 2753-2755.
[14]	霍兆义，尹洪超，赵亮，都健. 国内换热网络综合方法研究进展与展望[J]. 化工进展, 2012, 31(04 ): 726-731.
[15]	王琦，邓思旭，刘晶冰，汪浩. 提高正极材料磷酸铁锂倍率性能的研究进展[J]. 化工进展, 2011, 30(12): 2652-.