考虑未来功率需求的锂离子电池SOC多步预测  

SOC multi-step prediction of lithium-ion battery considering future power demand

作　　者：陈瑞[1,2] 陈俐 CHEN Rui;CHEN Li(Laboratory of Marine Power Plant and Automation,Shanghai Jiao Tong University,Shanghai 200240,China;State Key Laboratory of Ocean Engineering,Shanghai 200240,China)

机构地区：[1]上海交通大学动力装置与自动化研究所,上海200240 [2]海洋工程国家重点实验室,上海200240

出　　处：《电源技术》2024年第10期2013-2021,共9页Chinese Journal of Power Sources

基　　金：商飞-上海交通大学联合研究基金(CASEF-2022-M02)。

摘　　要：为提高荷电状态(SOC)多步预测精度,提出了基于长短期记忆(LSTM)的编码器-解码器用于SOC多步预测,在输入中考虑未来电池功率序列,在编码器和解码器上依次提取历史特征序列和未来功率序列的时间依赖信息。以某全电动飞机用锂离子电池包为应用案例,采集电池实验平台测试数据构建训练集和测试集,通过五折交叉验证选择模型的超参数。预测时长为300 s时,平均绝对误差、最大绝对误差和均方根误差分别为0.4231%、2.4847%和0.6450%。与没有输入未来功率的SOC多步预测模型进行对比,验证了在输入中增加未来功率能有效提高预测精度,与同样输入所有特征的多层感知机进行对比,验证了LSTM编码器-解码器具有更好的预测性能。In order to improve the accuracy of multi-step prediction of SOC,the encoder-decoder based on long short-term memory(LSTM)for the multi-step prediction of SOC was proposed.The future battery power sequence was considered in the input,and the time dependency data from the historical features sequence and future power sequence were extracted sequentially in the encoder and decoder.Taking a lithium-ion battery pack for an all-electric aircraft as an application case,the data of the battery experiment platform were collected to build the training set and the test set.The hyperparameters of the model were selected through 5-fold cross validation.When the prediction length is 300 s,the average absolute error,maximum absolute error and root mean square error are 0.4231%,2.4847%and 0.6450%,respectively.Compared with the SOC multi-step prediction model without input of future power,it’s verified that adding future power into the input can effectively improve the prediction accuracy.Compared with a multi-layer perceptron with the input of all features,it’s verified that the LSTMencoder-decoder has better prediction performance.

关 键 词：锂离子电池 SOC多步预测 长短期记忆 编码器-解码器 未来功率序列 

分 类 号：TM912.9[电气工程—电力电子与电力传动]