储能科学与技术

• XXXX •    

基于电化学阻抗谱与HO-TFRNN模型的磷酸铁锂电池SOC估计

江屹峰(), 全惠敏(), 卢继武   

  1. 湖南大学，湖南 长沙 410000
  • 收稿日期:2025-08-04 修回日期:2025-10-11
  • 通讯作者: 全惠敏 E-mail:923165050@qq.com;hmquan@hnu.edu.cn
  • 作者简介:江屹峰（2001—），性别男，硕士研究生，研究方向为锂离子电池状态估计，E-mail： 923165050@qq.com；
  • 基金资助:
    湖南省自然科学基金（2023JJ30146 湖南省自科）

Estimation of SOC in Lithium Iron Phosphate Batteries Based on Electrochemical Impedance Spectroscopy and HO-TFRNN Model

Yifeng Jiang(), Huimin Quan(), Jiwu Lu   

  1. Hunan University, Changsha 410000, Hunan, China
  • Received:2025-08-04 Revised:2025-10-11
  • Contact: Huimin Quan E-mail:923165050@qq.com;hmquan@hnu.edu.cn

摘要：

准确估算锂电池的荷电状态(State of Charge, SOC)对提升电池系统性能与可靠性至关重要。由于传统估算方法在进行磷酸铁锂电池SOC估计时，往往存在精度不足、稳定性差等问题，为了提高对电池SOC估计的精度，本文结合电化学阻抗谱（Electrochemical Impedance Spectroscopy,EIS）具有准确反映电池内部电化学状态的优势和混合时间-频率循环神经网络（Time-frequency Recurrent Neural Network, TFRNN）强大的序列处理能力，提出了一种基于EIS的和TFRNN的磷酸铁锂电池SOC估计算法。首先，本文在25℃环境下测量并分析了商用18650磷酸铁锂电池在不同SOC下的EIS曲线，通过等效电路建模与频域分析相结合的方式提取特征参数，并最终选取了0.1Hz处幅值与相位、100Hz处实部与虚部等参数作为SOC估计的输入特征构建数据集。然后，本文提出利用Transformer改进的TFRNN模型来估计电池SOC，同时引入河马仿生优化算法（Hippopotamus Optimization，HO）优化超参数以减少调参时间。利用自建的数据集进行实验验证，结果表明：该模型估算误差在3%以内，均方根误差（RMSE）为0.65050%，平均绝对误差（MAE）为0.66734%，最大误差（ME）为2.92306%。同时，本文还采集了20℃和15℃的数据集对结果进行对比验证，结果证明所提模型在多个数据集下都具有高度稳定性与精确性，能实现对磷酸铁锂电池SOC的精准估算。

关键词: 锂离子电池, 荷电状态, 电化学阻抗谱

Abstract:

Accurate estimation of the State of Charge (SOC) of lithium batteries is crucial for enhancing battery system performance and reliability. As traditional estimation methods often suffer from insufficient accuracy and poor stability when applied to lithium iron phosphate battery SOC estimation, this paper proposes an SOC estimation algorithm for lithium iron phosphate batteries based on Electrochemical Impedance Spectroscopy (EIS) and a Time-frequency Recurrent Neural Network (TFRNN). This approach combines the advantage of EIS in accurately reflecting the internal electrochemical state of the battery with the powerful sequence processing capability of the hybrid time-frequency recurrent neural network.First, the EIS curves of a commercial 18650 lithium iron phosphate battery at different SOC levels were measured and analyzed at 25 ℃. Feature parameters were extracted through a combination of equivalent circuit modeling and frequency domain analysis. Parameters such as magnitude and phase at 0.1 Hz, and real and imaginary parts at 100 Hz, were selected as input features for SOC estimation to construct the dataset.Subsequently, this paper proposes an improved TFRNN model enhanced with Transformer for battery SOC estimation, while introducing the Hippopotamus Optimization (HO) algorithm to optimize hyperparameters and reduce tuning time. Experimental validation using the self-constructed dataset demonstrated that the model achieves an estimation error within 3%, with a Root Mean Square Error (RMSE) of 0.65050%, a Mean Absolute Error (MAE) of 0.66734%, and a Maximum Error (ME) of 2.92306%.Additionally, datasets collected at 20 ℃ and 15 ℃ were used for comparative validation. The results confirm that the proposed model exhibits high stability and accuracy across multiple datasets, enabling precise estimation of the SOC of lithium iron phosphate batteries.

Key words: Lithium-ion battery, State of Charge, Electrochemical Impedance Spectroscopy