中国机械工程 ›› 2026, Vol. 37 ›› Issue (10): 2492-2501.DOI: 10.3969/j.issn.1004-132X.2026.10.012

• 系统能量管理与回收利用 • 上一篇    下一篇

考虑储能工况的退役动力电池退化建模与剩余寿命预测

马峰1,2(), 王海磊1,3, 黄顺虎1,3, 张华2,3   

  1. 1.冶金装备及其控制教育部重点实验室(武汉科技大学), 武汉, 430081
    2.湖北隆中实验室, 襄阳, 441000
    3.机械传动与制造工程湖北省重点实验室(武汉科技大学), 武汉, 430081
  • 收稿日期:2025-10-17 出版日期:2026-10-25 发布日期:2026-11-10
  • 通讯作者: 马峰
  • 作者简介:马峰*(通信作者),男,1989年生,副教授、博士研究生导师。研究方向为绿色低碳制造、资源高值化利用。E-mail:MF0425@163.com。
  • 基金资助:
    湖北省技术创新计划(2024BCB093);湖北省技术创新计划(2024EIA049);湖北省科技重大专项(2023BCA006)

Degradation Modeling and Remaining Useful Life Prediction of Retired Power Batteries Considering Energy Storage Conditions

MA Feng1,2(), WANG Hailei1,3, HUANG Shunhu1,3, ZHANG Hua2,3   

  1. 1.Key Laboratory of Metallurgical Equipment and Control Technology,Ministry of Education,Wuhan University of Science and Technology,Wuhan,430081
    2.Hubei Longzhong Laboratory,Xiangyang,Hubei,441000
    3.Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering,Wuhan University of Science and Technology,Wuhan,430081
  • Received:2025-10-17 Online:2026-10-25 Published:2026-11-10
  • Contact: MA Feng

摘要:

提出了一种融合鲸鱼优化算法(WOA)、卷积神经网络(CNN)、门控循环单元(GRU)和自注意力机制(SA)的深度学习模型,用于精准预测退役电池的剩余使用寿命。该模型充分发挥了CNN的特征提取能力、GRU的时序建模优势以及自注意力机制的长程依赖捕捉能力,并通过WOA实现超参数的全局最优配置。基于CALCE公开数据集的对比实验结果表明,所提WOA-CNN-GRU-SA模型在测试集上取得均方根误差(RMSE)为0.809 27%、平均绝对误差(MAE)为0.554 40%、决定系数R²为99.840%的预测性能,相较于次优模型WOA-CNN-LSTM-SA,在RMSE与MAE上分别降低10.35%与13.39%,体现出更高的预测精度与鲁棒性。研究结果可为退役动力电池的精准分级与储能系统的高效可靠管理提供关键技术支撑。

关键词: 退役动力电池, 剩余使用寿命, 储能系统, 深度学习, 梯次利用

Abstract:

A deep learning model integrating the whale optimization algorithm (WOA), convolutional neural network (CNN), gated recurrent unit (GRU), and self-attention(SA) mechanism was proposed to achieve accurate prediction of the RUL of retired batteries. The proposed model leveraged CNN for feature extraction, GRU for temporal dependency modeling, and SA for capturing long-range dependencies, while WOA was employed to globally optimize hyperparameters. Comparative experiments on the public CALCE dataset demonstrate that the proposed WOA-CNN-GRU-SA model achieves RMSE of 0.809 27%, an MAE of 0.554 40%, and an R² of 99.840% on the test set. Compared with the second-best model(WOA-CNN-LSTM-SA), the RMSE and MAE are reduced by 10.35% and 13.39%, respectively, indicating superior prediction accuracy and robustness. These results provide key technical support for accurate health grading of retired power batteries and efficient, reliable management of energy storage systems.

Key words: retired power battery, remaining useful life(RUL), energy storage system, deep learning, cascade utilization

中图分类号: