Adaptive hyperparameter-optimized transformer algorithm for lithium-ion battery SOH prediction

DU Xiongchun; DENG Dongliang

doi:10.19799/j.cnki.2095-4239.2026.0474

您当前的位置：

首页 >

文章列表页 >

Adaptive hyperparameter-optimized transformer algorithm for lithium-ion battery SOH prediction

Energy Storage Test\: Methods and Evaluation | 更新时间：2026-07-13

- Adaptive hyperparameter-optimized transformer algorithm for lithium-ion battery SOH prediction
- Energy Storage Science and Technology Vol. 15, Issue 6, Pages: 2392-2394(2026)
- 作者机构：
  
  三峡电能有限公司，湖北武汉 430021
- 作者简介：
- 基金信息：
- DOI：10.19799/j.cnki.2095-4239.2026.0474
  CLC： TM 911
- Received：01 June 2026，
  
  Published：28 June 2026
- 稿件说明：
移动端阅览
杜雄春, 邓栋梁. 基于超参数自适应优化的Transformer锂离子电池SOH预测算法[J]. 储能科学与技术, 2026, 15(6): 2392-2394.

DU Xiongchun, DENG Dongliang. Adaptive hyperparameter-optimized transformer algorithm for lithium-ion battery SOH prediction[J]. Energy Storage Science and Technology, 2026, 15(6): 2392-2394.
杜雄春, 邓栋梁. 基于超参数自适应优化的Transformer锂离子电池SOH预测算法[J]. 储能科学与技术, 2026, 15(6): 2392-2394. DOI： 10.19799/j.cnki.2095-4239.2026.0474.

DU Xiongchun, DENG Dongliang. Adaptive hyperparameter-optimized transformer algorithm for lithium-ion battery SOH prediction[J]. Energy Storage Science and Technology, 2026, 15(6): 2392-2394. DOI： 10.19799/j.cnki.2095-4239.2026.0474.

摘要

锂离子电池健康状态（SOH）的精准预测对储能系统的安全运行与运维决策至关重要。本文提出一种基于遗传算法（GA）自适应优化Transformer

超参数的SOH预测方法。首先，基于二阶RC等效电路模型提取端电压与工作电流作为输入特征，结合SOH构成多维时序输入；其次，利用Transformer的自注意力机制捕捉充放电过程中的长程依赖关系；最后，通过GA对学习率、批次大小、网络深度及注意力头数等关键超参数进行全局寻优，提升模型的泛化能力与预测精度。在NASA锂电池老化数据集上的实验结果表明，GA-Transformer在MAE、RMSE、MAPE和

²四项指标上均优于CNN、RNN、LSTM和GRU等对比模型，验证了所提方法的有效性与优越性。

Abstract

This paper presents an adaptive Transformer with GA-based hyperparameter tuning for battery SOH prediction in energy storage systems. First

terminal voltage and operating current are extracted as input features based on a second-order RC equivalent circuit model

combined with SOH to form multi-dimensional time-series inputs. Then

the self-attention mechanism of the Transformer is employed to capture long-range dependencies in the charging and discharging process. Finally

GA is utilized to globally optimize key hyperparameters including learning rate

batch size

network depth

and the number of attention heads

thereby enhancing the model's generalization capability and prediction accuracy. Experimental results on the NASA battery aging dataset demonstrate that the proposed GA-Transformer outperforms CNN

RNN

LSTM

and GRU in terms of MAE

RMSE

MAPE

and

validating the effectiveness and superiority of the proposed method.

关键词

Keywords

references

王宁, 刘晓峰, 陈泽华. 锂离子电池寿命预测综述[J]. 电器与能效管理技术, 2018(11): 1-13.

WANG N, LIU X F, CHEN Z H. Survey on lithium-ion battery life estimation[J]. Electrical & Energy Management Technology, 2018(11): 1-13.

杨淇, 孙桓五, 张凤博. 锂电SOC改进无迹卡尔曼滤波估算算法研究[J]. 机械设计与制造, 2021(10): 220-224.

YANG Q, SUN H W, ZHANG F B. Research on SOC improved unscented Kalman filter estimation algorithm for lithium batteries[J]. Machinery Design & Manufacture, 2021(10): 220-224.

VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Advances in Neural Information Processing Systems. Red Hook: Curran Associates, 2017: 5998-6008.

邓涛, 罗卫兴, 李志飞, 等. 双卡尔曼滤波法估计电动汽车电池健康状态[J]. 电池, 2018, 48(2): 95-99. DOI:10.19535/j.1001-1579.2018.02.007.

DENG T, LUO W X, LI Z F, et al. Estimation state of health of electric vehicle battery by dual Kalman filter[J]. Battery Bimonthly, 2018, 48(2): 95-99. DOI:10.19535/j.1001-1579.2018.02.007.

SHEN S Q, LIU B C, ZHANG K, et al. Toward fast and accurate SOH prediction for lithium-ion batteries[J]. IEEE Transactions on Energy Conversion, 2021, 36(3): 2036-2046. DOI:10.1109/TEC.2021.3052504.

高铭琨, 徐海亮, 吴明铂. 基于等效电路模型的动力电池SOC估计方法综述[J]. 电气工程学报, 2021, 16(1): 90-102. DOI:10.11985/2021.01.013.

GAO M K, XU H L, WU M B. Review of SOC estimation methods for power battery based on equivalent circuit model[J]. Journal of Electrical Engineering, 2021, 16(1): 90-102. DOI:10.11985/2021.01.013.

吴铁洲, 刘思哲, 张晓星, 等. 基于FA-BP神经网络的锂离子电池SOH估算[J]. 电池, 2021, 51(1): 21-25. DOI:10.19535/j.1001-1579.2021.01.006.

李敏强. 遗传算法的基本理论与应用[M]. 北京: 科学出版社, 2002.LI M Q. Basic theory and application of genetic algorithm[M]. Beijing: Science Press, 2002.

YAYAN U, ARSLAN A T, YUCEL H. A novel method for SoH prediction of batteries based on stacked LSTM with quick charge data[J]. Applied Artificial Intelligence, 2021, 35(6): 421-439. DOI:10.1080/08839514.2021.1901033.

张少宇, 伍春晖, 熊文渊. 采用门控循环神经网络估计锂离子电池健康状态[J]. 红外与激光工程, 2021, 50(2): 20200339. DOI:10.3788/IRLA20200339.

ZHANG S Y, WU C H, XIONG W Y. State of health estimation for lithium-ion batteries using recurrent neural networks with gated recurrent unit[J]. Infrared and Laser Engineering, 2021, 50(2): 20200339. DOI:10.3788/IRLA20200339.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

State of health estimation for lithium batteries based on short-term random charging data and optimized convolutional neural network

Accelerated life prediction of lithium-ion batteries using data-driven approaches

Estimating lithium-ion battery health using automatic feature extraction and channel attention mechanisms for multi-timescale modeling

State-of-health estimation of lithium-ion batteries based on multifeature analysis and LSTM-XGBoost model

A SOH estimation model for energy storage batteries based on multiple cycle features

Related Author

SHEN Jiangwei

SHE Yixin

SHU Xing

LIU Yonggang

WEI Fuxing

XIA Xuelei

CHEN Zheng

TIAN Chengwen

Related Institution

School of Transportation Engineering, Kunming University of Science and Technology

Key Laboratory of Advanced Manufacture Technology for Automobile Parts (Chongqing University of Technology) Ministry of Education

College of Mechanical Engineering, Chongqing University

National Active Distribution Network Technology Research Center

. of Vehicular Multi-Energy Drive Systems, Ministry of Education, Beijing

⁰