基于DOD-LN-GPR模型的锂离子电池SOH估计方法  

SOH ESTIMATION METHOD OF LITHIUM-ION BATTERIES BASED ON DOD-LN-GPR MODEL

作　　者：黄佳茵 白俊琦 贤燕华[1] Huang Jiayin;Bai Junqi;Xian Yanhua(School of Electronics and Information Engineering/School of Integrated Circuits,Guangxi Normal University,Guilin 541006,China)

机构地区：[1]广西师范大学电子与信息工程学院/集成电路学院,桂林541006

出　　处：《太阳能学报》2025年第2期60-69,共10页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(6197022931)。

摘　　要：针对锂离子电池健康状态(SOH)的估计中预测精度不高、健康特征输入冗余、数据预处理繁琐的问题,提出一种基于放电深度(DOD)的改进高斯过程回归SOH预测模型。在锂离子电池的放电曲线中,计算出锂离子电池的放电深度,并将其作为唯一的健康特征。同时改进传统的高斯过程回归(GPR)算法,利用线性(LIN)和神经网络(NN)的组合核函数(LIN+NN)拟合锂离子电池容量全局衰退和局部波动的趋势,从而建立DOD-LN-GPR锂离子电池SOH估计模型。在NASA数据集中,首先进行不同核函数的实验比对,验证所提组合核函数预测精度的优势;其次,通过减小训练集与测试集比例,证明所提估计方法在少量训练样本上仍能有较好的预测效果;最后,将所提DOD-LN-GPR模型在不同训练集下与其他SOH估计模型进行对比,结果表明该模型具有较好的精度和鲁棒性。To address the issues of low prediction accuracy,redundant health feature inputs,and cumbersome data preprocessing in the estimation of the State of Health(SOH)of Lithium-ion batteries,an improved SOH prediction model of Gaussian process regression based on depth of discharge(DOD)is proposed.In the discharge curve of the lithium battery,the discharge depth of the lithium-ion battery is calculated and taken as the only health feature.At the same time,the traditional Gaussian process regression(GPR)algorithm is improved,and the combined kernel function(LIN+NN)of linear(LIN)and neural network(NN)is used to fit the trend of global decline and local fluctuation of lithium-ion battery capacity,and the DOD-LN-GPR lithium-ion battery SOH estimation model is established.In the NASA data set,the experimental comparison of different kernel functions is carried out to verify the superiority of the prediction accuracy of the proposed combined kernel functions.Then,by reducing the ratio of training set to test set,it is proved that the proposed estimation method can still have good prediction effect on a small number of training samples.Finally,the proposed DOD-LN-GPR model is compared with other SOH estimation models under different training sets,and the results show that the model has better accuracy and robustness.

关 键 词：锂离子电池 状态估计 电池管理系统 高斯过程回归 放电深度 

分 类 号：TK02[动力工程及工程热物理]