基于SE-SAE特征融合和BiLSTM的锂电池寿命预测  

Life prediction of lithium battery based on SE-SAE feature fusion and BiLSTM

作　　者：叶震 李琨[1] 李梦男 高宏宇 YE Zhen;LI Kun;LI Mengnan;GAO Hongyu(Faculty of Information Engineering and Automation,Kunming University of Technology,Kunming Yunnan 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500

出　　处：《电源技术》2023年第6期745-749,共5页Chinese Journal of Power Sources

基　　金：国家自然科学基金(82160787)。

摘　　要：预测锂电池剩余使用寿命(RUL)时,针对电池外部特性参量电流、电压等单一的健康因子(HI)对电池退化特性无法完整覆盖的问题,提出一种结合通道注意力机制(SENet)和栈式自编码(SAE)进行特征融合并引入双向长短期记忆(BiLSTM)实现锂电池RUL的预测方法。充分提取锂电池电压、电流等HI。利用SAE对多个锂电池HI特征进行特征融合,并结合SENet通道注意力机制,增加重要特征在提取过程中的表现能力。利用BiLSTM网络对融合HI进行训练预测。采用NASA和马里兰大学计算机辅助寿命周期工程中心(CALCE)锂电池数据集进行验证,训练预测数据均采用50%的比例划分,预测结果的均方根误差(RMSE)平均值达到0.017。When predicting the remaining useful life(RUL)of lithium batteries,a prediction method combining channel attention mechanism(SENet)and the stacked auto encoder(SAE)for feature fusion and introducing directional long short-term memory(BiLSTM)was proposed to solve the problem that single health indicator(HI),such as current and voltage,could not fully cover the degradation characteristics of batteries.HI was fully extracted,such as voltage and current of lithium batteries.SAE was used for feature fusion of multiple HI features of lithium battery,and SENet channel attention mechanism was used to enhance the expressive ability of important features in the extraction process.BiLSTM network was used to train and predict fusion HI.The validation was conducted by using the lithium battery dataset of NASA and the computer-aided life cycle engineering(CALCE)of University of Maryland.The training and prediction data were divided into 50%proportions,and the average RMSE(root mean square error)of the prediction results reaches 0.017.

关 键 词：SENet 栈式自编码 特征融合 双向长短期记忆网络 电池寿命预测 

分 类 号：TM912[电气工程—电力电子与电力传动]