基于深度学习的锂离子电池剩余寿命估计  

Remaining Useful Life Estimation for Lithium-ion Battery Using Deep Learning Method

作　　者：曹孟达 张涛[1] 王羽 张亚军 刘亚杰 CAO Mengda;ZHANG Tao;WANG Yu;ZHANG Yajun;LIU Yajie(College of Systems Engineering,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]国防科技大学系统工程学院,湖南长沙410073

出　　处：《无线电工程》2021年第7期641-648,共8页Radio Engineering

基　　金：面向应急任务的XXX电源系统保障能力评估。

摘　　要：锂离子电池剩余使用寿命(Remaining Useful Life,RUL)的准确预测能够帮助更好地了解锂离子电池的工作状态,人工智能和深度学习的发展为锂离子电池RUL预测提供了新的数据驱动方法。基于自动编码器和深度神经网络,提出了一种深度学习框架的锂离子电池RUL预测方法。通过自动编码器对锂离子电池进行退化特征的提取和融合,以融合特征作为输入、电池容量作为输出,训练深度神经网络用于多电池RUL预测。将该方法应用于NASA的锂离子电池的真实数据集。实验结果表明,该深度学习预测框架中自动编码器提取融合特征后能够提高RUL预测精度,预测方法具有很好的可靠性和准确性。The accurate prediction of the remaining useful life(RUL)of lithium-ion batteries can help to better understand the working state of lithium-ion batteries.With the development of artificial intelligence and deep learning,new data-driven methods are provided for lithium-ion battery RUL prediction.An automatic encoder and deep neural network are combined to propose a deep learning framework for lithium ion battery RUL prediction.First,the automatic encoder is used to extract and fuse the degraded features of lithium-ion batteries.Then,the deep neural network is trained with fusion features as input and capacity as output for multi-cell RUL prediction.This method is applied to NASA's real data set of lithium ion battery cycle life.The experimental results show that the automatic encoder in the deep learning prediction framework can improve the RUL prediction accuracy after extracting the fusion features.The prediction method has good reliability and accuracy.

关 键 词：锂离子电池 剩余寿命估计 深度学习 自动编码器 深度神经网络 

分 类 号：TM911[电气工程—电力电子与电力传动]