基于SSA-LSTM网络模型的锂离子电池健康状态预测  

State-of-health prediction of lithium-ion batteries based on SSA-LSTM network model

作　　者：俞寅森 朱涛[2] 位承君[2] 叶杨倩 廖强强 付在国 YU Yinsen;ZHU Tao;WEI Chengjun;YE Yangqian;LIAO Qiangqiang;FU Zaiguo(School of Environmental and Chemical Engineering,Shanghai University of Electric Power,Shanghai 200090,China;Xi′an Thermal Power Research Institute Co.,Ltd.,Xi′an 710054,Shaanxi Province,China)

机构地区：[1]上海电力大学环境与化学工程学院,上海200090 [2]西安热工研究院有限公司,陕西西安710054

出　　处：《化学工程》2024年第12期14-20,57,共8页Chemical Engineering(China)

基　　金：上海市科委项目(19DZ2271100)。

摘　　要：锂离子电池被广泛用作各种设备的电源,因此对锂离子电池S_(OH)(健康状态)的快速准确预测是降低电池故障的重要手段。由于LSTM(长短期记忆)网络可以从时间序列中找出变量变化的特征、趋势以及发展规律,进而对变量的未来变化进行有效地预测,因此已成为预测锂离子电池S_(OH)的一种流行的深度学习网络方法。未优化超参数的LSTM方法很容易导致电池S_(OH)预测模型的精度低。针对锂离子电池S_(OH)预测问题,提出一种基于SSA(麻雀搜索算法)优化LSTM的方法。提取一种新的健康指标-充电电压PDF(概率密度函数)曲线峰值处的峰度,并将其用作S_(OH)预测模型的输入,以实现对电池S_(OH)的准确预测。实验结果表明,SSA优化的LSTM模型的预测精度优于未优化模式。当训练集仅占总数据的20%时,NCA(镍钴铝)电池S_(OH)预测结果的均方根误差E_(RMSE)在0.7%以内,最大绝对误差<2.0%。SSA-LSTM可以在训练数据有限的情况下准确预测电池S_(OH)。Lithium-ion batteries are widely used as power sources for various devices,so rapid and accurate prediction of the S_(OH)(state of health)of lithium-ion batteries is an important means to reduce battery failures.Due to its ability to identify the characteristics,trends and development patterns of variable changes from time series,LSTM(long short-term memory)network is a popular deep learning network method for predicting the future changes of lithium-ion battery S_(OH).The LSTM method without optimizing hyperparameters can easily lead to low accuracy in battery S_(OH) prediction models.A method based on SSA(sparrow search algorithm)was proposed to optimize LSTM for the prediction of S_(OH) in lithium-ion batteries.A new health indicator-the kurtosis at the peak of the charging voltage PDF(probability density function)curve was extracted and used as input for the S_(OH) prediction model to achieve accurate prediction of battery S_(OH).The experimental results show that the prediction accuracy of the LSTM model optimized by SSA is better than that of the unoptimized model.When the training set only accounts for 20%of the total data,the root mean square error E_(RMSE) of the NCA(nickel cobalt aluminum)battery S_(OH) prediction results is within 0.7%,and the maximum absolute error is less than 2.0%.SSA-LSTM can accurately predict battery S_(OH) under limited training data.

关 键 词：锂离子电池 健康状态 麻雀搜索算法 长短期记忆网络 超参数 峰度 

分 类 号：TM912[电气工程—电力电子与电力传动]