储能科学与技术 ›› 2025, Vol. 14 ›› Issue (3): 1258-1269.doi: 10.19799/j.cnki.2095-4239.2024.1124

• 储能新锐科学家专刊 • 上一篇    下一篇

一种基于ICA-T特征和CNN-LA-BiLSTM的锂离子电池健康状态估计方法

张朝龙1,2(), 陈阳1, 刘梦玲1, 张俣峰1, 华国庆1, 阴盼昐1   

  1. 1.金陵科技学院智能科学与控制工程学院,江苏 南京 211169
    2.武汉大学电气与自动化学院,湖北 武汉 430072
  • 收稿日期:2024-11-27 修回日期:2024-12-13 出版日期:2025-03-28 发布日期:2025-04-28
  • 通讯作者: 张朝龙 E-mail:zhangchaolong@126.com
  • 作者简介:张朝龙(1982—),男,博士,教授,研究方向为锂离子电池智能管理,E-mail: zhangchaolong@126.com
  • 基金资助:
    国家重点研发计划“智能电网技术与装备”专项(2023YFB2406900);江苏省高等学校基础科学 (自然科学)研究重大项目(23KJA480002);江苏高校“青蓝工程”中青年学术带头人,金陵科技学院高层次人才资助项目(jit-rcyj-202202);2024年大学生创新训练项目(202413573009Z);2024年金陵科技学院“科教融合”项目(2024KJRH09)

A state of health estimation method for lithium-ion batteries using ICA-T features and CNN-LA-BiLSTM

Chaolong ZHANG1,2(), Yang CHEN1, Mengling LIU1, Yufeng ZHANG1, Guoqing HUA1, Panpan YIN1   

  1. 1.College of Intelligent Science and Control Engineering, Jinling Institute of Technology, Nanjing 211169, Jiangsu, China
    2.School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, Hubei, China
  • Received:2024-11-27 Revised:2024-12-13 Online:2025-03-28 Published:2025-04-28
  • Contact: Chaolong ZHANG E-mail:zhangchaolong@126.com

摘要:

为了解决锂离子电池健康状态(SOH)估计精度不足以及退化过程描述不准确的问题,本文提出了一种基于卷积神经网络-局部注意力-双向长短期记忆神经网络(CNN-LA-BiLSTM)的锂离子电池SOH估计方法。首先,测量锂离子电池在充电阶段的充电时间、电流、电压、容量以及温度等数据。然后,对锂离子电池进行增量容量分析,提取增量容量(IC)曲线的面积作为锂离子电池的电特征;计算锂离子电池充电阶段的温度积分,作为温度特征;将曲线面积与温度相结合,用作锂离子电池SOH估计的联合特征增量容量面积-温度(ICA-T)。随后,利用CNN-LA-BiLSTM方法建立SOH估计模型,在模型中,引入局部注意力(LA)优化卷积神经网络(CNN)的权重和偏差,使用Huber损失函数优化模型参数从而获得良好的SOH估计效果。利用本实验室的2组锂离子电池数据开展测试,结果表明,提出的方法能有效地估计电池的SOH,平均绝对百分比误差(MAPE)为0.5794%,均方根误差(RMSE)为0.0099,决定系数(R2)为0.9961。与传统方法相比,本文提出的方法在电池SOH估计中表现出了更优的性能。

关键词: 锂离子电池, 健康状态估计, 卷积神经网络-局部注意力-双向长短期记忆神经网络, 增量容量, Huber损失函数

Abstract:

To address the challenges of insufficient estimation accuracy and inaccurate degradation modeling of lithium-ion battery state of health (SOH), this study proposes a lithium-ion battery SOH estimation method based on a convolutional neural network-local attention-bidirectional long short-term memory (CNN-LA-BiLSTM) model. First, the charging time, current, voltage, capacity, and temperature of the lithium-ion battery are measured during the charging phase. The lithium-ion battery undergoes incremental capacity (IC) analysis, and the IC curve area is extracted as an electrical characteristic of the lithium-ion battery. The temperature integral during charging the lithium-ion battery is calculated as a temperature characteristic. These features are combined into a joint IC area-temperature metric for SOH estimation of lithium-ion batteries. Then, the CNN-LA-BiLSTM model is constructed, incorporating LA to optimize the weights and biases of the CNN, while Huber loss function is used to optimize model parameters for enhanced SOH estimation accuracy. Results show that the proposed method effectively estimates the SOH of the battery, achieving a mean absolute percentage error of 0.5794%, root mean square difference of 0.0099, and a coefficient of determination of 0.9961. Compared with traditional methods, the proposed method shows better performance in battery SOH estimation.

Key words: lithium-ion battery, SOH, CNN-LA-BiLSTM, IC, Huber loss function

中图分类号: