摘要
在锂离子电池的应用中,快速充电技术的普及带来了显著的便利,但不同快速充电策略对电池健康状态(SOH)的影响显著。在快速充电条件下,锂离子电池SOH的准确估计面临着鲁棒性差、计算成本高等困难。该文在多级恒流快充场景中,利用充电电压和放电电压曲线提取等电压范围采样计数作为健康特征双输入估计锂离子电池SOH。同时,采用多任务学习框架,以应对部分健康特征缺失的情况,增强模型的鲁棒性。在大型公开快充数据集上进行验证,结果显示,在多级恒流快充条件下,锂离子电池SOH估计的方均根误差和平均绝对误差均在1%以内,决定系数在0.98以上,且表现出较强的鲁棒性。
In the application of lithium-ion batteries(LIBs),the popularity of fast charging technology has brought significant convenience to consumers,particularly for portable devices such as electric vehicles and smartphones,with an increasing demand for fast charging.However,fast charging strategies have a significant impact on the state of health(SOH)of LIBs,particularly in a multistage constant-current fast charging environment.Although fast charging improves charging efficiency,battery chemistry and thermal management issues accelerate the battery's aging due to the drastic changes in charging current and voltage,which affect its performance and lifetime.Therefore,it becomes imperative to accurately estimate the SOH of Li-ion batteries under such charging conditions.Existing SOH estimation methods suffer from insufficient robustness and high computational cost under fast charging conditions.Building on the previous innovative method,this study utilizes the charging and discharging voltage profiles to extract the sampling counts within the equal voltage range as the dual inputs for the health features.This method leverages critical information from the battery charging and discharging process,thereby significantly reducing computational complexity by simplifying the feature extraction process.A multi-task learning framework is used to enhance the robustness of the SOH estimation further.The framework utilizes the shared layer of the LSTM model to share information among multiple related tasks,thereby optimizing the learning process and performance,especially in cases where some health features are missing.In this way,the model can better adapt to the uncertainty and noise in the data,thereby improving its performance under fast charging strategies.A large public fast charging dataset is used for extensive testing.The results demonstrate that the proposed method performs well under multistage constant-current fast-charging conditions,with the root mean square error(RMSE)and mean absolute error(MAE)of the SOH estimation within 1%.The coefficient of determination(R2)reaches more than 0.98.In addition,the proposed method demonstrates strong robustness and stability when handling various charging strategies and missing features.In summary,the voltage profile-based SOH estimation method proposed in this study provides a new solution for the health management of Li-ion batteries under fast charging conditions,which applies to fast charging scenarios of electric vehicles and portable devices.It provides essential theoretical support for the design,optimization,and management of LIBs in the future.
作者
毛玲
林涛
赵建辉
赵晋斌
Mao Ling;Lin Tao;Zhao Jianhui;Zhao Jinbin(College of Electrical Engineering Shanghai University of Electric Power,Shanghai 200090 China;Engineering Research Center of Offshore Wind Technology Ministry of Education Shanghai University of Electric Power,Shanghai 200090 China)
出处
《电工技术学报》
北大核心
2026年第2期714-724,共11页
Transactions of China Electrotechnical Society
关键词
锂离子电池
健康状态
多任务学习
长短期记忆网络
Lithium-ion battery
state of health
multi-task learning framework
long short-term memory
