储能科学与技术 ›› 2024, Vol. 13 ›› Issue (10): 3630-3641.doi: 10.19799/j.cnki.2095-4239.2024.0398

• 储能测试与评价 • 上一篇    

基于递归图多尺度特征的储能锂离子电池剩余寿命预测方法

谢毓广1(), 李金中1, 邹文豪2(), 毛磊2()   

  1. 1.国网安徽省电力有限公司电力科学研究院,安徽 合肥 230601
    2.中国科学技术大学工程科学学院,安徽 合肥 230026
  • 收稿日期:2024-05-06 修回日期:2024-06-06 出版日期:2024-10-28 发布日期:2024-10-30
  • 通讯作者: 邹文豪,毛磊 E-mail:95662479@qq.com;wenhaozou@mail.ustc.edu.cn;leimao82@ustc.edu.cn
  • 作者简介:谢毓广(1979—),男,博士,正高级工程师,研究方向为电力系统优化运行、新能源并网及储能技术,E-mail:95662479@qq.com
  • 基金资助:
    国家电网有限公司科技项目(SGAHDK00DJJS2200281);安徽省自然科学基金能源互联网联合基金(2208085UD03)

Prediction of remaining useful life for lithium-ion battery using recurrence plot analysis

Yuguang XIE1(), Jinzhong LI1, Wenhao ZOU2(), Lei MAO2()   

  1. 1.State Grid Anhui Electric Power Research Institute, Hefei 230601, Anhui, China
    2.School of Engineering Science, University of Science and Technology of China, Hefei 230026, Anhui, China
  • Received:2024-05-06 Revised:2024-06-06 Online:2024-10-28 Published:2024-10-30
  • Contact: Wenhao ZOU, Lei MAO E-mail:95662479@qq.com;wenhaozou@mail.ustc.edu.cn;leimao82@ustc.edu.cn

摘要:

锂离子电池在长期循环使用过程中不可避免地会出现性能退化,这直接影响储能锂离子电池系统的稳定运行。为此,本工作提出基于递归图多尺度特征的锂离子电池剩余寿命预测方法,用于解决从一维状态信号中提取关键退化特征的局限性。鉴于递归图像内蕴含丰富的时空退化特征,首先,构建深度学习多尺度特征提取架构,通过可变大小的感受野,识别同一电压区域在多个周期内的时间维度变化以及相邻电压区域之间空间维度的递归图像时空演变,进而提取深层多尺度特征,用于实现从多尺度特征到RUL的映射建模。再次,通过开展综合评估实验,对所提出方法的预测效果进行系统验证。研究结果表明,该方法使用有限数量的充电过程递归图作为输入,能实现模型快速收敛和准确预测。此外,在跨倍率预测场景中,相较于浅层指标,可实现在2C倍率下将绝对误差和均方根误差的指标性能提升约7倍和5.7倍。最后,通过开展与一维序列输入的对比实验,进一步验证基于递归图多尺度特征进行锂离子电池剩余寿命预测的有效性,实现了各评价指标约50%和43%的性能提升,同时成像所需时序电压采样点数据量相对较小。

关键词: 锂离子电池, 剩余寿命预测, 递归图, 多尺度特征提取

Abstract:

During the long-term cycling of lithium-ion batteries (LIBs), performance degradation is inevitable, which directly impacts the stable operation of the battery system. To address this, this paper proposes a novel for predicting the remaining useful life of LIBs based on multi-scale features derived from recurrence plots. This approach aims to overcome the limitations of extracting key degradation features from one-dimensional (1D) signals, such as voltage data. By leveraging the rich spatiotemporal degradation features captured in recurrence plots, we first develop a deep learning architecture for effective multi-scale feature extraction. This architecture detects temporal variations within the same voltage region across multiple cycles and examines the spatial evolution of recurrence plots between adjacent voltage regions using variable-sized receptive fields. This approach enables the extraction of deep multi-scale features, which are then mapped to RUL modeling. Comprehensive evaluation experiments were performed to systematically validate the predictive effectiveness of the proposed method. The results suggest that by using a limited number of charge process recurrence plots as input, the model achieves rapid convergence and accurate predictions. Additionally, in cross-rate prediction scenarios, the proposed method improves performance metrics, with absolute error (MAE) and root mean squared error (RMSE) reduced by approximately 7-fold and 5.7-fold, respectively, at a 2C rate compared to shallow indicators. Finally, comparative experiments with 1D sequence inputs further validate the effectiveness of predicting the RUL of LIBs using multiscale features of recurrence plots. This approach achieves performance improvements of approximately 50% and 43% in various evaluation metrics, while requiring relatively minimal time-series voltage sampling data for imaging.

Key words: lithium-ion batteries, remaining useful life prediction, recurrence plot, multi-scale feature extraction

中图分类号: