几种机器学习算法的锂电池SOC估计研究

Research on SOC estimation of lithium batteries using several machine learning algorithms

荷电状态(SOC)是锂电池管理系统的重要参数,对其能否准确估计关系到电池系统的优化和使用寿命。为了消除传统方法在SOC估计精度、实时性等方面的不足,提出了一种改进的估计锂电池SOC的LightGBM算法。在Panasonic-18650PF-Data公开数据集的基础上应用其进行锂电池SOC估计,减小数据训练误差,并评估机器学习模型的泛化误差。将随机森林、支持向量机、线性回归及神经网络等常用估计模型算法作为对比。数据表明,改进的LightGBM算法估计时间为随机森林的1/22、支持向量机的1/88及神经网络的1/1330,估计误差低于0.06。研究表明,该方法初步解决了现有锂电池SOC估计方法速率慢、拟合率低的问题,并拓展了该算法新的应用场景。

It is well known that state of charge(SOC)is one of crucial parameters for lithium-ion battery management system,and the optimization and service life of the battery system is related to the estimation accurate of SOC.Considering the defect regarding accuracy and real-time performance of SOC estimation using traditional methods,an improved LightGBM algorithm has been proposed to estimate the SOC of lithium battery.In this study,the LightGBM algorithm is used to estimate lithium battery SOC based on Panasonic-18650P-Data set,which has been done to reduce the data training error and evaluate generalization errors of machine learning models.Moreover,four kinds of common estimation models including random forest,support vector machine,linear regression and neural network are compared with the LightGBM algorithm,and simulations show that the estimation time consumed by LightGBM algorithm is equal to 1/22 of random forest,1/88 of support vector machine and 1/1330 of neural network.Simultaneously,the estimation error is less than 0.06.Research conducted in this paper indicates that the method can solve approvingly the problem about slow speed and low fitting rate of the current SOC estimation methods for lithium batteries,and it is natural to expand the potential application for LightGBM algorithm.