基于注意力机制和CNN-LSTM融合模型的锂电池SOC预测  

SOC Prediction for Lithium Battery Based on Fusion Model of Attention Mechanism and CNN-LSTM

作　　者：张帅涛 蒋品群[1] 宋树祥[1] 夏海英[1] ZHANG Shuaitao;Student Member;CPSS;JIANG Pinqun;SONG Shuxiang;XIA Haiying(School of Electronic and Information Engineering/School of Integrated Circuits,Guangxi Normal University,Guilin 541004,China)

机构地区：[1]广西师范大学电子与信息工程学院/集成电路学院,桂林541004

出　　处：《电源学报》2024年第5期269-277,共9页Journal of Power Supply

基　　金：广西科技重大专项资助项目(AA20302003,AA23023010)。

摘　　要：为提高锂电池荷电状态SOC(state-of-charge)预测精度,提出1种基于注意力机制和卷积神经网络-长短时记忆CNN-LSTM(convolution neural network-long short-term memory)融合模型的锂电池荷电状态预测方法。该模型采用一维CNN和LSTM神经网络学习得到SOC与锂电池放电数据的非线性关系,以及SOC序列存在的长期依赖性。同时,该模型采用“多对一”的结构,将当前时刻的锂电池SOC与多个历史时刻的放电数据建立映射关系,并通过注意力机制关注到对当前时刻SOC影响较大的历史放电数据,进一步提升SOC的预测准确度。动态工况下的锂电池SOC预测实验表明,该方法在不同温度条件下的平均预测误差为0.89%,与SVM、GRU和XGBoost相比,分别降低了81.2%、66.7%和56.5%,且优于未融合注意力机制的LSTM和CNN-LSTM,具有较高的预测精度和应用价值。To improve the state-of-charge(SOC)prediction accuracy of lithium battery,a prediction method based on the fusion model of Attention mechanism and convolution neural network-long short-term memory(CNN-LSTM)is proposed.This model uses one-dimensional CNN and LSTM neural network to learn the nonlinear relationship between SOC and lithium battery discharge data,as well as the long-term dependence existing in SOC sequences.At the same time,it adopts a“many-to-one”structure and establishes a mapping relationship between the SOC at the present moment and the discharge data at multiple historical moments,and pays attention to the historical discharge data which has a greater influence on the SOC at the present moment through the Attention mechanism,thus further improving the SOC prediction accuracy.The SOC prediction experiments under dynamic conditions show that the average prediction error of the proposed method is 0.89%under different temperature conditions,which is 81.2%,66.7%and 56.5%lower than those of SVM,GRU and XGBoost algorithms,respectively.In addition,this method is also superior to LSTM and CNNLSTM models that do not combine the Attention mechanism,showing a higher prediction accuracy and higher application values.

关 键 词：锂电池 荷电状态 卷积神经网络 长短时记忆神经网络 注意力机制 

分 类 号：TM912[电气工程—电力电子与电力传动] U482.3[交通运输工程—载运工具运用工程]