基于1DCNN-LSTM的锂离子电池SOH预测  

Hybrid 1DCNN-LSTM model for predicting lithium ion battery state of health

作　　者：王英楷 张红[1] 王星辉 WANG Yingkai;ZHANG Hong;WANG Xinghui(College of Physics and Information Engineering FuZhou University,Fuzhou 350116,Fujian,China)

机构地区：[1]福州大学物理与信息工程学院,福建福州350116

出　　处：《储能科学与技术》2022年第1期240-245,共6页Energy Storage Science and Technology

基　　金：国家自然科学基金(83417013);福建省科技厅(2019J06008,2018J01535)。

摘　　要：为了提高锂离子电池健康状态(SOH)的预测精准度和稳定性,针对常规特征选取复杂且无法有效利用等问题,提出了一种联合一维卷积(1DCNN)与长短记忆网络(LSTM)的电池SOH预测方法。首先采用多通道串联电压、电流、温度构建多维特征,然后采用1DCNN从样本数据中提取高级数据特征输入LSTM中以有效利用历史信息,最后通过全连接层输出电池SOH的预测结果。采用NASA锂离子电池容量衰减数据,对所应用的联合算法进行验证,结果表明,相较于其他预测算法,基于1DCNN-LSTM的算法具有更准确的SOH预测结果,其平均绝对误差(MAE)为0.01左右,且失效点误差周期(RUL)小于2个周期。To improve the prediction accuracy and stability of the lithium-ion battery state of health(SOH),a battery SOH prediction method combining one-dimensional convolution(1DCNN)and long and short-term memory network(LSTM)is proposed herein to solve the problems of the complex selection of conventional features and the inability to effectively use them.First,multichannel series voltage,current,and temperature are used to construct multi-dimensional features.Second,1DCNN is used to extract advanced data features from the sample data and input them into the LSTM to effectively utilize historical information.Finally,the SOH prediction results of the battery are output through a full connection layer.The combined algorithm is verified by the capacity attenuation data of the NASA lithium battery.The results show that compared with other prediction algorithms,the algorithm based on the 1DCNN-LSTM has a more accurate prediction result of the SOH with a mean absolute error of 0.01 and a failure point error of less than two cycles.

关 键 词：1DCNN LSTM 锂电池 多通道特征 电池寿命 

分 类 号：TM912[电气工程—电力电子与电力传动]