储能科学与技术 ›› 2023, Vol. 12 ›› Issue (2): 552-559.doi: 10.19799/j.cnki.2095-4239.2022.0574

• 储能测试与评价 • 上一篇    下一篇

基于BP-UKF算法的锂离子电池SOC估计

杨帆1(), 和嘉睿2, 陆鸣1, 陆玲霞1, 于淼2()   

  1. 1.浙江大学工程师学院,浙江 杭州 310015
    2.浙江大学电气工程学院,浙江 杭州 310027
  • 收稿日期:2022-10-08 修回日期:2022-11-08 出版日期:2023-02-05 发布日期:2023-02-24
  • 通讯作者: 于淼 E-mail:22060233@zju.edu.cn;zjuyumiao@zju.edu.cn
  • 作者简介:杨帆(1998—),男,硕士研究生,研究方向为锂离子电池SOC估计与均衡控制,E-mail:22060233@zju.edu.cn
  • 基金资助:
    中央高校基本科研业务费专项资金(226-2022-00164)

SOC estimation of lithium-ion batteries based on BP-UKF algorithm

Fan YANG1(), Jiarui HE2, Ming LU1, Lingxia LU1, Miao YU2()   

  1. 1.Polytechnic Institute, Zhejiang University, Hangzhou 310015, Zhejiang, China
    2.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, Zhejiang, China
  • Received:2022-10-08 Revised:2022-11-08 Online:2023-02-05 Published:2023-02-24
  • Contact: Miao YU E-mail:22060233@zju.edu.cn;zjuyumiao@zju.edu.cn

摘要:

电池的荷电状态(state of charge,SOC)是电池管理的重要指标之一,准确的SOC估计是保证锂离子电池安全有效运行的必要条件。为提高锂离子电池SOC估计的准确性,本文基于二阶Thevenin等效模型,提出一种将无迹卡尔曼滤波(unscented Kalman filter,UKF)与BP(back propagation)神经网络相结合的SOC估计方法。在通过混合功率脉冲特性测试获取模型参数的基础上,首先利用UKF算法对电池SOC进行初步估计,通过非线性点变换的方法避免了扩展卡尔曼滤波(extended Kalman filter,EKF)在线性化过程中对系统造成的精度损失;其次,构建三层BP神经网络,综合考虑锂离子电池的充放电电压、电流等参数,对估计结果进行修正,将估计误差从初始估计结果中排除,以达到更加准确的估计结果。通过电池充放电测试仪采集锂离子电池在动态应力测试下的充放电数据,并在不同的噪声环境下将本文提出的BP-UKF算法与EFK算法和UKF算法进行对比实验分析。实验结果表明,本文提出的BP-UKF算法的最大误差在2.18%以内,平均误差在0.54%以内,均方根误差在0.0044以内,较EKF算法和UKF算法有较大程度地提升;并且在较大的环境噪声条件下,BP-UKF算法的准确性提升更为明显。

关键词: SOC估计, 无迹卡尔曼滤波算法, 锂离子电池, 二阶Thevenin模型, BP神经网络

Abstract:

The state of charge (SOC) of batteries is one of the most important indicators for battery management. An accurate SOC estimation is necessary to ensure the safe and effective operations of lithium-ion batteries. To improve the accuracy of the SOC estimation of lithium-ion batteries, this paper proposes a SOC estimation method by adopting the integration of the unscented Kalman filter (UKF) and the back propagation (BP) neural network, based on the second-order Thevenin equivalent model.By obtaining the model parameters through hybrid pulse power characteristic tests, the UKF algorithm is used to estimate the initial SOC of the battery. The nonlinear point transformation method is used to avoid the accuracy loss caused by the system linearization process in the extended Kalman filter (EKF). Then, a three-layer BP neural network is constructed, and the estimation errors are corrected by considering the charging and discharging voltage and current, along with other parameters of lithium-ion batteries. The estimation errors are then excluded from the initial estimation results to achieve more accurate results. The charging and discharging results of lithium-ion batteries in the dynamic stress test were collected by the battery charging and discharging tester. The BP-UKF algorithm proposed in this paper was compared with the EKF algorithm and the traditional UKF algorithm under different noise environments. The experimental results show that the maximum error of the proposed BP-UKF algorithm is within 2.18%, the mean absolute percentage error is within 0.54%, and the root mean square error is within 0.0044, demonstrating evident improvements compared with the other two algorithms. In addition, the accuracy of the BP-UKF algorithm is improved more significantly under the condition of large environmental noises.

Key words: SOC estimation, unscented Kalman filter algorithm, lithium-ion battery, second-order Thevenin model, BP neural network

中图分类号: