基于DeepAR与特征选择的锂离子电池在线状态估计  

Online State Estimation of Lithium-ion Batteries Based on DeepAR and Feature Selection

作　　者：史永胜[1] 任嘉睿 李锦 张凯[1] SHI Yongsheng;REN Jiarui;LI Jin;ZHANG Kai(School of Electrical and Control Engineering,Shaanxi University of Science and Technology,Xi’an 710021,China)

机构地区：[1]陕西科技大学电气与控制工程学院,西安710021

出　　处：《电源学报》2023年第2期163-171,共9页Journal of Power Supply

基　　金：国家自然科学基金资助项目(61871259);陕西省重点研发计划资助项目(2021-GY135)。

摘　　要：电池健康状态SOH(state-of-health)和荷电状态SOC(state-of-charge)估计是电池管理系统的核心功能。目前,状态估计存在依赖大量历史数据以及单一状态估计适应性差的问题,因此提出一种基于DeepAR与特征选择的锂离子电池状态估计模型。首先,提取电池恒流充电过程中电压、温度及时间间隔数据,组成3组老化特征作为模型输入,用于估计SOH;然后,在估计SOC时考虑SOH估计值,消除了电池老化因素对SOC估算的负面影响;最后,在不同工况下的牛津电池数据集上进行实验验证,并与其他两种算法模型进行误差与收敛性对比。结果表明,所提模型在冷启动估计方面具有较强的优势,SOH和SOC估计精度较高。The estimation of state-of-health(SOH)and state-of-charge(SOC)is the core function of a battery management system.At present,the state estimation depends on a large number of historical data,and the single-state estimation has a poor adaptability.To solve these problems,a lithium-ion battery state estimation model based on DeepAR and feature selection is proposed.First,the voltage,temperature and time interval data in the constant-current charging process of a battery are extracted to form three groups of aging characteristics as the model input for predicting SOH.Then,the SOH estimation is considered in SOC prediction to eliminate the negative effect of battery aging on SOC estimation.Finally,experimental verification was carried out on the Oxford battery data set under different working conditions,and the error and convergence of the proposed model were compared with those of the other two algorithms.Results show that the proposed model has a strong advantage in terms of cold boot estimation,and its estimation accuracy of SOH and SOC is higher.

关 键 词：锂离子电池 健康状态 荷电状态 自回归循环神经网络 

分 类 号：TM912[电气工程—电力电子与电力传动]