发电技术 ›› 2025, Vol. 46 ›› Issue (6): 1154-1163.DOI: 10.12096/j.2096-4528.pgt.24068

• 储能 • 上一篇    

基于改进自适应蜜獾算法优化时间卷积网络的车载锂离子电池健康状态估计

张效伟1, 衣振晓2,3,4, 王凯2   

  1. 1.中海油石化工程有限公司,山东省 济南市 250101
    2.青岛大学电气工程学院,山东省 青岛市 266071
    3.青岛大学威海创新研究院,山东省 威海市 264200
    4.山东索想智能科技有限公司,山东省 潍坊市 261101
  • 收稿日期:2024-05-13 修回日期:2024-07-20 出版日期:2025-12-31 发布日期:2025-12-25
  • 通讯作者: 王凯
  • 作者简介:张效伟(1988),女,硕士,工程师,研究方向为太阳能热发电技术,420997393@qq.com
    衣振晓(1997),男,博士研究生,研究方向为储能技术、新型储能元件、健康状态估计、剩余使用寿命预测,1824164415@qq.com
    王凯(1985),男,教授,博士,研究方向为储能技术、新型储能元件、健康状态估计、剩余使用寿命预测,本文通信作者,wkwj888@163.com
  • 基金资助:
    国家自然科学基金项目(52037005)

State of Health Estimation of On-Board Lithium-Ion Batteries Using Temporal Convolutional Network Optimized by Improved Self-Adaptive Honey Badger Algorithm

Xiaowei ZHANG1, Zhenxiao YI2,3,4, Kai WANG2   

  1. 1.CNOOC Petrochemical Engineering Company Limited, Jinan 250101, Shandong Province, China
    2.College of Electrical Engineering, Qingdao University, Qingdao 266071, Shandong Province, China
    3.Weihai Innovation Research Institute, Qingdao University, Weihai 264200, Shandong Province, China
    4.Shandong Suoxiang Intelligent Technology Company Limited, Weifang 261101, Shandong Province, China
  • Received:2024-05-13 Revised:2024-07-20 Published:2025-12-31 Online:2025-12-25
  • Contact: Kai WANG
  • Supported by:
    National Natural Science Foundation of China(52037005)

摘要:

目的 锂离子电池作为新能源汽车的重要动力来源,准确的健康状态(state of health,SOH)估计对于设计安全可靠的汽车电池管理系统至关重要。传统方法普遍存在忽略容量恢复及特征有效性不足等问题,严重影响估算精度。为此,提出了一种虑及电池容量恢复的锂离子电池SOH估算方法。 方法 将中值绝对偏差与Savitzky-Golay滤波相结合,在数据预处理阶段有效去除异常值和噪声以提高特征的有效性,然后进行特征分解,去除冗余信息,减轻模型计算负担。将高度相关的特征作为时间卷积网络模型的输入,降低数据维度并减轻神经网络计算复杂度。提出了一种改进的自适应蜜獾算法以优化网络的超参数,加快模型收敛并提高网络性能。 结果 所提方法具有较高的准确性,均方根误差和平均绝对误差均低于0.007。 结论 所提方法具有较高的鲁棒性,能够对车载锂离子电池SOH进行有效估计,并能满足实际应用需求。

关键词: 储能, 新能源, 电动汽车, 锂离子电池, 健康状态估计, 改进蜜獾算法, 时间卷积网络, 数据驱动, 荷电状态

Abstract:

Objectives Lithium-ion batteries, as an important power source for new energy vehicles, require accurate state of health (SOH) estimation to design safe and reliable battery management systems. Traditional methods often overlook issues such as capacity recovery and insufficient feature effectiveness, which significantly affects estimation accuracy. To address these issues, a novel SOH estimation method for lithium-ion batteries that considers battery capacity recovery is proposed. Methods By combining the median absolute deviation with the Savitzky-Golay filter in the data preprocessing stage, the model effectively removes outliers and noise to improve the effectiveness of the features. Subsequently, feature decomposition is performed to remove redundant information and alleviate the computational load of the model. Highly correlated features are then selected as inputs for the temporal convolutional network model, reducing data dimensionality and simplifying the computational complexity of the neural network. Furthermore, an improved self-adaptive honey badger algorithm is proposed to optimize the hyperparameters of the network, accelerating model convergence and enhancing network performance. Results The proposed method has a high level of accuracy, with both the root mean squared error and the mean absolute error being lower than 0.007. Conclusions The proposed method exhibits high robustness, enabling effective SOH estimation of on-board lithium-ion batteries and meeting requirements of practical application.

Key words: energy storage, new energy, electric vehicles, lithium-ion battery, state of health estimation, improved honey badger algorithm, temporal convolutional network, data-driven, state of charge

中图分类号: