储能科学与技术 ›› 2020, Vol. 9 ›› Issue (1): 131-137.doi: 10.19799/j.cnki.2095-4239.2019.0189

• 储能系统与工程 • 上一篇    下一篇

基于高斯过程回归的锂离子电池SOC估计

李嘉波(), 魏孟, 叶敏, 焦生杰, 徐信芯   

  1. 长安大学公路养护装备国家工程实验室,陕西 西安 710064
  • 收稿日期:2019-08-25 修回日期:2019-10-15 出版日期:2020-01-05 发布日期:2019-10-24
  • 作者简介:李嘉波(1992—),男,博士研究生,研究方向为新能源方向,E-mail:431991454@qq.com
  • 基金资助:
    国家自然科学基金青年项目(51805041)

SOC estimation of lithium-ion batteries based on Gaussprocess regression

Jiabo LI(), Meng WEI, Min YE, Shengjie JIAO, Xinxin XU   

  1. Highway Maintenance Equipment National Engineering Laboratory, Chang'an University, Xi'an 710064, Shaanxi, China
  • Received:2019-08-25 Revised:2019-10-15 Online:2020-01-05 Published:2019-10-24

摘要:

电池状态估计(SOC)在电池管理系统(BMS)尤为重要,由于SOC估计易受温度、荷载、充放电效率等外界因素的影响,因此估计精度很难保证。目前,有很多国内外学者利用机器学习算法进行SOC估计,然而神经网络(NN)的估计精度依赖于样本个数,支持向量机(SVM)在参数寻优时已陷入局部最优。因此为了提高SOC的估计精度,提出了基于高斯过程回归(GPR)的锂离子电池在线的估计方法,根据电池的测量参数,包括电流、电压、温度作为GPR模型的输入,SOC作为模型的输出,进行模型训练,并利用梯度下降法进行参数寻优。通过仿真和恒流充放电实验采集的数据来验证模型的有效性,并与SVM、LSSVM和NN相比,验证了模型的有效性和高精度性。

关键词: SOC, 高斯过程回归, 锂离子电池

Abstract:

Battery state estimation (state of charge, SOC) is particularly important in a battery management system (BMS). It is difficult to ensure its accuracy because SOC estimation is vulnerable to temperature, load, charging and discharging efficiency, and other external factors. Currently, many scholars use machine learning algorithms to estimate the SOC. However, the estimation accuracy of a neural network (NN) is dependent on the number of samples. The support vector machine (SVM) falls into a local optimum in parameter optimization. An online estimation method is proposed based on Gaussian process regression (GPR) for lithium-ion batteries to improve the estimation accuracy of the SOC. Based on the battery measurement parameters, including current, voltage, and temperature, as input to the GPR model and the SOC as an output of the model, the model is trained, and the parameters are optimized using the gradient descent method. The validity of the model is verified by simulation and data collected from the constant current charging and discharging experiments. Compared with the SVM, LSSVM, and NN, the validity and feasibility of the model are verified.

Key words: SOC, Gauss process regression, lithium-ion battery

中图分类号: