基于Transformer的锂离子电池SOC估计方法  

Lithium-ion Battery SOC Estimation Method Based on the Transformer

作　　者：李杰 康龙云[1,2] 岳睿 谢缔 LI Jie;KANG Long-yun;YUE Rui;XIE Di(South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学,电力学院,广东广州510640 [2]华南理工大学,广东省绿色能源技术重点实验室,广东广州510640

出　　处：《电力电子技术》2023年第7期64-66,共3页Power Electronics

基　　金：广东省重点领域研发计划(2019B090911001)。

摘　　要：此处提出了基于Transformer的荷电状态(SOC)预测模型以提高对锂离子电池的SOC预测精度及效率。首先将易于测量的电压电流等数据作为编码器的输入,利用编码器端的多头注意力机制来提取深层特征,充分地利用输入数据的特征信息,同时将目标SOC作为解码器端的输入,将编码器的输出输入至解码器端,最后输出时移后的SOC得到预测结果。为了防止标签泄露,在解码器端的输入采用了掩码机制,同时实现了卷积神经网络提取特征和循环神经网络利用数据变化的时序性。利用锂电池的工况放电实验数据进行训练,以均方误差(MSE)作为评价标准。实验结果表明,所提出的基于Transformer模型具有较高的预测精度。In order to improve the accuracy and efficiency of predicting the state of charge of lithium-ion batteries,a state of charge(SOC)prediction model is proposed based on Transformer.First of all,the easy-to-measure voltage and current data as the input of the encoder,the use of the multi-head attention mechanism at the encoder end to extract deep features,the full use of the characteristic information of the input data,while the target SOC as the input of the decoder end,the output of the encoder input to the decoder end,and finally the output time-shifted SOC to get the prediction result,in order to prevent label leakage,the decoder end using a masking mechanism to handle the input.At the same time,the convolutional neural network extracts features and the recurrent neural network takes advantage of the timing of data changes.The experimental data of the working condition discharge of lithium batteries are used for training,and the mean squared error(MSE)is used as evaluation criteria.Experimental results show that the Tran-sformer-based model proposed has high prediction accuracy.

关 键 词：锂离子电池 均方误差 荷电状态 

分 类 号：TM911[电气工程—电力电子与电力传动]