储能科学与技术 ›› 2021, Vol. 10 ›› Issue (6): 2342-2351.doi: 10.19799/j.cnki.2095-4239.2021.0291

• 储能测试与评价 • 上一篇    下一篇

基于IBAS-NARX神经网络的锂电池荷电状态估计

曹新宇1(), 彭飞1(), 李立伟2, 尹建光3   

  1. 1.青岛大学电气工程学院,山东 青岛 266071
    2.青岛大学威海创新研究院,山东 威海 264229
    3.国网山东省电力公司电力科学研究院,山东 济南 250002
  • 收稿日期:2021-06-26 修回日期:2021-07-23 出版日期:2021-11-05 发布日期:2021-11-03
  • 作者简介:曹新宇(1996—),男,硕士研究生,研究方向为新能源汽车电控系统开发等,E-mail:402141904@qq.com|彭飞,讲师,主要研究方向为新能源动力系统状态估计与故障诊断,E-mail:kilmer_pf@126.com
  • 基金资助:
    国网山东省电力公司科技项目(ZY-2021-09)

SOC estimation of lithium battery based on IBAS-NARX neural network model

Xinyu CAO1(), Fei PENG1(), Liwei LI2, Jianguang YIN3   

  1. 1.School of electrical engineering of Qingdao University, Qingdao 266071, Shandong, China
    2.Weihai Innovation Institute of Qingdao University, Weihai 264229, Shandong, China
    3.State Grid Shandong Electric Power Research Institute, Ji'nan 250002, Shandong, China
  • Received:2021-06-26 Revised:2021-07-23 Online:2021-11-05 Published:2021-11-03

摘要:

在使用神经网络方法估计锂电池荷电状态时,传统荷电状态适应度评价函数存在仅考虑均方误差等网络权值参数的缺点,忽略了拓扑参数对模型的影响。故本文提出在适应度评价函数设计中综合考虑输入/输出时序相关性、隐层神经元数量等模型拓扑参数和网络权值参数的加权影响,并将其引入带外部输入非线性自回归神经网络建模方法的锂电池荷电状态估计中,进而基于改进天牛须搜索算法实现了上述模型拓扑参数与网络权值参数的协同辨识优化。仿真结果表明,本文所提出方法能够提高多种复杂工况下的锂电池荷电状态估计精度,在DST标准工况和WLTC标准工况下锂电池荷电状态的均方根误差分别达到3.38×10-3和8.75×10-4,相比于未经改进的天牛须搜索算法优化NARX神经网络在均方根误差上估计精度分别提升了42.4%和20.5%。

关键词: 锂电池, 荷电状态估计, 天牛须搜索算法, NARX神经网络

Abstract:

When using neural network methods to estimate the state of charge of lithium batteries, the traditional state of charge fitness evaluation function has the disadvantage of only considering the network weight parameters such as the mean square error and ignoring the influence of topology parameters on the model. Therefore, this paper proposes to consider the weighted influence of model topology parameters and network weight parameters, such as input/output timing correlation, and the number of hidden layer neurons, in the design of the fitness evaluation function and introduce it into a nonlinear autoregressive neural network with external input. Based on the improved longhorn whisker search algorithm, the estimation of the state of charge of the lithium battery in the modeling method involves the collaborative identification and optimization of the above-mentioned model topology parameters and network weight parameters. The simulation results show that the method proposed in this paper can improve the accuracy of estimating the state of charge of a lithium battery under a variety of complex working conditions. The root mean square error of the lithium battery state of charge under DST standard working condition and WLTC standard working condition is 3.38×10-3 and 8.75×10-4, respectively, which improves the estimation accuracy of the root mean square error by 42.4% and 20.5% when compared to the BAS-NARX neural network.

Key words: lithium battery, SOC estimation, beetle antennae search algorithm, NARX neural network

中图分类号: