基于卡尔曼滤波算法优化Transformer模型的锂离子电池健康状态预测方法

Kalman filter optimize Transformer method for state of health prediction on lithium-ion battery

黄煜峰 ¹梁焕超 ¹许磊²

扫码查看

作者信息

1. 沈阳航空航天大学电子信息工程学院,辽宁沈阳 110136
2. 应急管理部沈阳消防研究所,辽宁沈阳 110034
折叠

摘要

回顾了锂离子电池健康状态(state of health,SOH)的预测方法,本文提出了一种基于卡尔曼滤波器(Kalman filter)优化的Transformer网络,利用历史数据预测电池SOH.首先,通过添加高斯噪声、自动编码器重构对电池数据进行预处理,去除电池数据中原始噪声并强化数据特征;其次,利用提出的KF-Transformer(Kalman filter-transformer)算法模型提取电池健康状态变化特征,使得Transformer网络能够更好地捕捉电池健康状态的非线性变化;最后,通过线性层完成电池健康状态变化特征到电池健康状态预测的映射,得到锂离子电池健康状态的预测结果.本文使用3个不同充放电策略、不同测试环境下的锂离子电池数据集[分别为美国国家航空航天局(NASA)数据集、马里兰大学CALCE CS2数据集和CALCE CX2数据集]来验证本文提出的锂离子电池健康状态预测算法的鲁棒性和准确性,且本文算法对于SOH不同状态、不同循环次数的预测均具有较好的结果.研究结果显示,本文方法的平均绝对百分比误差(mean absolute percentage error,MAPE)能控制在2%,决定系数(R-square,R2)为0.987,并与多层感知机(MLP,multilayer perceptron)、循环神经网络(RNN,recurrent neura network)、长短时记忆(LSTM,long short-term memory)、门控循环单元(GRU,gated recurrent unit)进行比较,证明了本文方法的优越性.

Abstract

Reviewing methods for predicting the lithium-ion battery state of health(SOH),we propose a battery SOH prediction method,which uses Transformer network based on Kalman Filter.Firstly,the battery data is preprocessed by adding Gaussian noise and auto-encoder reconstrution to remove the original noise from the data and strengthen the data features.Secondly,the battery data is extracted using the proposed(Kalman Filter-Transformer,KF-Transformer)model to extract the features of battery SOH changes,so that the Transformer network can better capture the nonlinear changes of battery SOH.Finally,the mapping from the features to the prediction of battery SOH is accomplished through the linear layer to obtain the prediction results of lithium-ion battery SOH.In this paper,three datasets(NASA,CALCE CS2 and CALCE CX2)are used for training and testing,and the two datasets use different temperatures and different batteries.The R2 value of the predicted result in this paper is 0.987 and the mean absolute percentage error(MAPE)value is 1.8%,which is better than Multilayer Perceptron(MLP),Recurrent Neural Network(RNN),Long Short-Term Memory(LSTM),Gated Recurrent Unit(GRU).These results demonstrate the accuracy and stability of the proposed method for SOH prediction.

关键词

锂离子电池/电池健康状态/自动编码器/Transformer网络/卡尔曼滤波器

Key words

lithium-ion battery/battery state of health/auto-encoder/transformer network/Kalman Filter

引用本文复制引用

基金项目

沈阳市自然科学基金专项(23-503-6-18)

辽宁省科技厅应用基础研究计划项目(2023JH2/101300145)

出版年

2024

储能科学与技术

化学工业出版社

储能科学与技术

CSTPCD北大核心

影响因子：0.852

ISSN：2095-4239

参考文献量3

段落导航