基于NARX神经网络的车用锂离子电池SOH时间序列预测

徐东辉; 石本改; 徐丽琴; 叶雪强; 王丽娜

doi:10.3969/j.issn.1001-2222.2022.06.012

车用发动机 >

2022 , Vol. 0 >Issue 6: 71 - 75

DOI: https://doi.org/10.3969/j.issn.1001-2222.2022.06.012

栏目

基于NARX神经网络的车用锂离子电池SOH时间序列预测

徐东辉 ,
石本改 ,
徐丽琴 ,
叶雪强 ,
王丽娜

展开

（1.南昌师范学院数学与计算机科学系，江西南昌 330032；2.广东机电职业技术学院汽车学院，广东广州 510550；3.长沙理工大学汽车与机械工程学院，湖南长沙 410076）

收起

SOH Time-Series Prediction of Vehicle Lithium-Ion Battery Based on NARX Network

XU Donghui ,
SHI Bengai ,
XU Liqin ,
YE Xueqiang ,
WANG Lina

Expand

(1.Department of Mathematics and Computer Science,Nanchang Normal Universiy,Nanchang 330032,China;2.School of Automobile,Guangdong Vocational College of Mechanical and Electrical Technology,Guangzhou 510550,China;3.School of Automotive and Mechanical Engineering,Changsha University of Science and Technology,Changsha 410076,China)

Fold

摘要

锂离子电池模型参数具有慢时变特性，因而准确预测锂离子电池健康状态(state of health,SOH)存在较大的难题。利用非线性自回归（Nonlinear AutoRegressive with eXogenous input，NARX）神经网络建立了SOH时间序列预测模型，通过重构技术将预测模型的一维输入时间序列重构成多维状态空间，并且采用重构后的时间序列数据对NARX神经网络对进行训练，然后利用训练后的NARX神经网络进行预测得到最终的SOH时间序列预测值；试验结果显示，预测模型比RBF神经网络的均方误差提高了近6个百分点，收敛速度提高了近30 s，表明了基于NARX的SOH时间序列预测模型的预测精度及响应速度都较好。

关键词： 锂离子电池; 健康状态; 神经网络; 预测

本文引用格式

徐东辉 , 石本改 , 徐丽琴 , 叶雪强 , 王丽娜 . 基于NARX神经网络的车用锂离子电池SOH时间序列预测[J]. 车用发动机, 2022 , 0(6) : 71 -75 . DOI: 10.3969/j.issn.1001-2222.2022.06.012

Abstract

The state of health(SOH) of lithium ion battery was difficult to predict accurately due to the slow and time-varying characteristic of model parameters. The SOH time-series prediction model was hence built by using NARX(nonlinear autoregressive with exogenous input) neural network. The one-dimensional input time series of prediction model was reconstructed into the multidimensional state space by the reconstruction technology, the NARX neural network was trained with the reconstructed time-series data, and then the final SOH time-series prediction value was obtained by using the trained NARX neural network. The experimental results show that the mean square error and the convergence speed of the proposed model is nearly 6 percentage points and 30 seconds superior to those of RBF neural network, which indicates that the prediction accuracy and response speed of the SOH time-series prediction model based on NARX are both better.

Key words： lithium-ion battery; state of health; neural network; prediction

Options

摘要页面

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract