基于注意力机制和多任务LSTM的锂电池容量预测方法*

doi:10.11985/2022.04.006

电气工程学报 ›› 2022, Vol. 17 ›› Issue (4): 41-50.doi: 10.11985/2022.04.006

• 特邀专栏：电化学储能系统安全管理与运维 • 上一篇下一篇

扫码分享

基于注意力机制和多任务LSTM的锂电池容量预测方法^*

鲁南¹(), 欧阳权¹(), 黄俍卉²(), 王志胜¹()

1.南京航空航天大学自动化学院南京 211106
2.浙江科技学院自动化与电气工程学院杭州 310023

收稿日期:2022-05-15 修回日期:2022-06-24 出版日期:2022-12-25 发布日期:2023-02-03
通讯作者: 欧阳权，男，1991年生，博士，副教授，硕士研究生导师。主要研究方向为锂电池管理系统、新能源系统集成与控制。E-mail：ouyangquan@nuaa.edu.cn
作者简介:鲁南，男，2000年生。主要研究方向为锂电池管理系统。E-mail：nanlu@nuaa.edu.cn
黄俍卉，女，1991年生，博士，讲师。主要研究方向为新能源系统集成与控制、燃料电池建模与控制。E-mail：120036@zust.edu.cn
王志胜，男，1970年生，博士，教授，博士研究生导师。主要研究方向为智能机器人技术，锂电池管理系统。E-mail：wangzhisheng@nuaa.edu.cn
基金资助:
*国家自然科学基金(61903189);中国博士后科学基金(2020M681589);中央高校基本科研业务费(NS2021023);浙江省自然科学基金(LQ22F030010)

Capacity Prediction of Lithium-ion Batteries Based on Multi-task LSTM with Attention Mechanism

LU Nan¹(), OUYANG Quan¹(), HUANG Lianghui²(), WANG Zhisheng¹()

1. School of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 211106
2. School of Automation and Electrical Engineering, Zhejiang University of Science and Technology, Hangzhou 310023

Received:2022-05-15 Revised:2022-06-24 Online:2022-12-25 Published:2023-02-03
Contact: OUYANG Quan, E-mail：ouyangquan@nuaa.edu.cn

摘要/Abstract

摘要：

精确的锂电池容量预测可以有效降低电池失效带来的风险与损失。基于神经网络的时间序列预测模型是电池容量预测领域中十分常见的方法。但是，过去的模型预测大多只考虑了未来的目标点，而没有考虑目标点附近信息对预测目标的辅助作用。提出一种基于注意力机制和多任务LSTM的锂电池容量预测方法(MT-LSTM)，实现未来不同时刻信息的互补，提高预测的准确性，其中使用硬参数共享方法建立多个任务之间的联系，使用卷积神经网络提取不同抽象水平的特征。通过注意力机制与LSTM模型的结合，有效地提高电池容量预测精度。将所提出的MT-LSTM模型与传统神经网络进行对比，试验结果表明所提模型有更高的预测精度。此外为多任务学习与注意力机制设计了对比试验，验证了这两种方法对电车容量预测精度的积极影响。

关键词: 电池容量预测, 长短期记忆神经网络, 注意力机制, 多任务学习

Abstract:

Accurate capacity prediction of lithium-ion batteries can effectively reduce the risk and loss caused by battery failure. The time series prediction model based on neural network is a very common method in the field of battery capacity prediction. However, most of the predictions of the past models only considered the future target points, but did not consider the auxiliary role of the information near the target points. A capacity prediction method of lithium-ion batteries based on multi-task LSTM with attention mechanism is proposed to realize the complementation of information at different times in the future and improve the prediction accuracy. The hard parameter sharing method is used to establish the connection among multiple tasks, and the convolutional neural network is used to extract features at different levels of abstraction. Comparing with the traditional neural network, and the experimental results show that the proposed MT-LSTM model has higher prediction accuracy. In addition, comparison experiments are designed for the multi-task learning and the attention mechanism to verify the positive effects of these two methods on the prediction accuracy of battery capacity.

Key words: Battery capacity prediction, long short-term memory neural network, attention mechanism, multi-task learning

中图分类号:

TM561

鲁南, 欧阳权, 黄俍卉, 王志胜. 基于注意力机制和多任务LSTM的锂电池容量预测方法^*[J]. 电气工程学报, 2022, 17(4): 41-50.

LU Nan, OUYANG Quan, HUANG Lianghui, WANG Zhisheng. Capacity Prediction of Lithium-ion Batteries Based on Multi-task LSTM with Attention Mechanism[J]. Journal of Electrical Engineering, 2022, 17(4): 41-50.

图/表 12

图1

图2

图3

图4

图5

表1

图6

表2

表3

图7

表4

图8

参考文献 27

[1]	HAN Xuebing, LU Languang, ZHENG Yuejiu, et al. A review on the key issues of the lithium-ion battery degradation among the whole life cycle[J]. ETransportation, 2019, 1:100005. doi: 10.1016/j.etran.2019.100005
[2]	ZHANG Jingliang, LEE J. A review on prognostics and health monitoring of Li-ion battery[J]. Journal of Power Sources, 2011, 196(15):6007-6014. doi: 10.1016/j.jpowsour.2011.03.101
[3]	XIONG Rui, LI Linlin, TIAN Jinpeng. Towards a smarter battery management system:A critical review on battery state of health monitoring methods[J]. Journal of Power Sources, 2018, 405:18-29. doi: 10.1016/j.jpowsour.2018.10.019
[4]	刘勇智, 詹群, 盛增津, 等. 最小二乘支持向量机在航空蓄电池剩余容量预测中的应用[J]. 蓄电池, 2013, 50(3):118-120,144.
	LIU Yongzhi, ZHAN Qun, SHENG Zengjin, et al. Application of least-squares support vector machine for residual capacity prediction of aviation batteries[J]. Battery, 2013, 50(3):118-120,144.
[5]	ZHU Mingye, OUYANG Quan, WAN Yong, et al. Remaining useful life prediction of lithium-ion batteries:A hybrid approach of grey-Markov chain model and improved Gaussian process[J/OL]. IEEE Journal of Emerging and Selected Topics in Power Electronics,[2021-07-19],DOI:10.1109/JESTPE.2021.3098378. doi: 10.1109/JESTPE.2021.3098378
[6]	CHAOUI H, IBE-EKEOCHA C C. State of charge and state of health estimation for lithium batteries using recurrent neural networks[J]. IEEE Transactions on Vehicular Technology, 2017, 66(10):8773-8783. doi: 10.1109/TVT.2017.2715333
[7]	PARK K, CHOI Y, CHOI W, et al. LSTM-based battery remaining useful life prediction with multi-channel charging profiles[J]. IEEE Access, 2020, 8:20786-20798. doi: 10.1109/ACCESS.2020.2968939
[8]	CUI Shengming, JOE I. A dynamic spatial-temporal attention-based GRU model with healthy features for state-of-health estimation of lithium-ion batteries[J]. IEEE Access, 2021, 9:27374-27388. doi: 10.1109/ACCESS.2021.3058018
[9]	YOU G W, PARK S, OH D. Diagnosis of electric vehicle batteries using recurrent neural networks[J]. IEEE Transactions on Industrial Electronics, 2017, 64(6):4885-4893. doi: 10.1109/TIE.2017.2674593
[10]	REN Lei, DONG Jiabao, WANG Xiaokang, et al. A data-driven auto-CNN-LSTM prediction model for lithium-ion battery remaining useful life[J]. IEEE Transactions on Industrial Informatics, 2020, 17(5):3478-3487. doi: 10.1109/TII.2020.3008223
[11]	BAHDANAU D, CHO K, BENGIO Y. Neural machine translation by jointly learning to align and translate[J]. Computer Science, 2014:1409.0473.
[12]	RAFFEL C, ELLIS D P W. Feed-forward networks with attention can solve some long-term memory problems[P]. DOI:10.48550/arXiv.1512.08756,2015. doi: 10.48550/arXiv.1512.08756,2015
[13]	CHENG Jiezhu, HUANG Kaizhu, ZHENG Zibin. Towards better forecasting by fusing near and distant future visions[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(4):3593-3600. doi: 10.1609/aaai.v34i04.5766
[14]	CARUANA R. Multitask learning[J]. Machine Learning, 1997, 28(1):41-75. doi: 10.1023/A:1007379606734
[15]	SAHA B, GOEBEL K. Battery data set[DB]. NASA AMES Prognostics Data Repository, 2007. http://tiarc.nasa.gov/project/prognostic-datarepository.
[16]	HU Xiaosong, JIANG Jiuchun, CAO Dongpu, et al. Battery health prognosis for electric vehicles using sample entropy and sparse Bayesian predictive modeling[J]. IEEE Transactions on Industrial Electronics, 2015, 63(4):2645-2656.
[17]	REYES O, VENTURA S. Performing multi-target regression via a parameter sharing-based deep network[J]. International Journal of Neural Systems, 2019, 29(9):1950014. doi: 10.1142/S012906571950014X
[18]	LIBERMAN N, TROPE Y. The role of feasibility and desirability considerations in near and distant future decisions:A test of temporal construal theory[J]. Journal of Personality and Social Psychology, 1998, 75(1):5. doi: 10.1037/0022-3514.75.1.5
[19]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8):1735-1780. doi: 10.1162/neco.1997.9.8.1735 pmid: 9377276
[20]	KATTENBORN T, LEITLOFF J, SCHIEFER F, et al. Review on convolutional neural networks (CNN) in vegetation remote sensing[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 173:24-49. doi: 10.1016/j.isprsjprs.2020.12.010
[21]	SANTURKAR S, TSIPRAS D, ILYAS A, et al. How does batch normalization help optimization?[J]. Advances in Neural Information Processing Systems, 2018, 31:2488-2498.
[22]	LI Shuqing, JU Chuankun, LI Jianliang, et al. State-of-charge estimation of lithium-ion batteries in the battery degradation process based on recurrent neural network[J]. Energies, 2021, 14(2):306. doi: 10.3390/en14020306
[23]	曹广华, 赵中林, 许昀昊. 基于GRU的锂电池组健康状态预测研究[J]. 吉林大学学报, 2022, 40(2):181-187.
	CAO Guanghua, ZHAO Zhonglin, XU Yunhao. Research on health state prediction of lithium battery pack based on GRU[J]. Journal of Jilin University, 2022, 40(2):181-187.
[24]	张菁, 吴尚青. 基于LSTM和注意力机制的锂电池荷电状况预测[J]. 九江学院学报, 2021, 36(3):29-34.
	ZHANG Jing, WU Shangqing. Li-ion battery charge condition prediction based on LSTM and attention mechanism[J]. Journal of Jiujiang College, 2021, 36(3):29-34.
[25]	SAON S, HIYAMA T. Predicting remaining useful life of rotating machinery based artificial neural network[J]. Computers & Mathematics with Applications, 2010, 60(4):1078-1087. doi: 10.1016/j.camwa.2010.03.065
[26]	王钟毅, 姬晓, 左思. 基于BP神经网络的锂电池剩余寿命预测[J]. 汽车实用技术, 2021, 46(1):8-9.
	WANG Zhongyi, JI Xiao, ZUO Si. Residual life prediction of lithium battery based on BP neural network[J]. Automotive Practical Technology, 2021, 46(1):8-9.
[27]	付强, 王华伟. 基于多层LSTM的复杂系统剩余寿命智能预测[J]. 兵器装备工程学报, 2022, 43(1):161-169.
	FU Qiang, WANG Huawei. Intelligent prediction of remaining life of complex systems based on multilayer LSTM[J]. Journal of Arms and Equipment Engineering, 2022, 43(1):161-169.

预测间隔	模型	B0005		B0007		B0018
预测间隔	模型	MAE	MSE	MAE	MSE	MAE	MSE
T+10	MT-LSTM	0.015 4	3.72×10^-4	0.013 8	4.00×10^-4	0.018 0	5.30×10^-4
	GRU	0.027 6	9.02×10^-4	0.014 9	3.78×10^-4	0.030 8	12.2×10^-4
	LSTM	0.035 7	14.6×10^-4	0.018 1	4.60×10^-4	0.031 7	13.1×10^-4
	ANN	0.024 9	8.67×10^-4	0.016 0	4.86×10^-4	0.034 0	14.6×10^-4
T+15	MT-LSTM	0.020 7	6.32×10^-4	0.019 7	6.35×10^-4	0.019 6	6.08×10^-4
	GRU	0.035 0	14.8×10^-4	0.017 6	4.79×10^-4	0.044 4	27.2×10^-4
	LSTM	0.051 2	28.4×10^-4	0.021 4	6.55×10^-4	0.048 1	32.1×10^-4
	ANN	0.021 6	6.59×10^-4	0.015 6	4.50×10^-4	0.044 8	27.1×10^-4

预测间隔	模型	B0005		B0007		B0018
预测间隔	模型	MAE	MSE	MAE	MSE	MAE	MSE
T+10	MT-LSTM	0.015 4	3.72×10^-4	0.013 8	4.00×10^-4	0.018 0	5.30×10^-4
	LSTM-CNN-LSTM	0.017 3	4.24×10^-4	0.014 2	4.12×10^-4	0.025 2	9.67×10^-4
	Deep LSTM	0.039 9	18.0×10^-4	0.021 9	6.24×10^-4	0.030 8	12.0×10^-4
T+15	MT-LSTM	0.020 7	6.32×10^-4	0.019 7	6.35×10^-4	0.019 6	6.08×10^-4
	LSTM-CNN-LSTM	0.020 1	6.46×10^-4	0.019 1	4.79×10^-4	0.023 9	9.39×10^-4
	Deep LSTM	0.056 2	34.3×10^-4	0.022 3	7.12×10^-4	0.047 3	31.6×10^-4

超参数组合	结合注意力机制		无注意力机制
超参数组合	MAE	MSE	MAE	MSE
H=16, m=16	0.022 8	6.69×10^-4	0.022 9	6.68×10^-4
H=32, m=32	0.021 3	9.12×10^-4	0.068 2	72.7×10^-4
H=64, m=64	0.015 4	3.72×10^-4	0.037 7	25.0×10^-4
H=16, m=64	0.026 0	7.86×10^-4	0.029 5	16.7×10^-4
均值	0.021 4	6.84×10^-4	0.039 5	26.35×10^-4

基于注意力机制和多任务LSTM的锂电池容量预测方法^*

Capacity Prediction of Lithium-ion Batteries Based on Multi-task LSTM with Attention Mechanism

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 27

相关文章 3

编辑推荐

Metrics

本文评价

[1]	蔡鑫祥, 撖奥洋, 周生奇, 菅学辉, 张智晟. 基于改进Bagging算法与模糊MP-LSTM融合的短期负荷预测模型^*[J]. 电气工程学报, 2022, 17(1): 164-170.
[2]	熊瑞, 闫良基, 王榘. 功率需求驱动的电动载运装备用动力电池充放电能力预测方法[J]. 机械工程学报, 2021, 57(20): 161-171.
[3]	董绍江, 裴雪武, 吴文亮, 汤宝平, 赵兴新. 基于多层降噪技术及改进卷积神经网络的滚动轴承故障诊断方法[J]. 机械工程学报, 2021, 57(1): 148-156.