锂离子电池健康状态的DCAE-Transformer预测方法研究

Method of DCAE-Transformer Prediction for the Health State of Lithium-Ion Batteries

李浩平 ¹于波涛 ¹孟荣华 ¹金朱鸿 ¹杜昕毅 ¹李景瑞¹

扫码查看

作者信息

1. 三峡大学机械与动力学院,湖北宜昌 443002
折叠

摘要

提出了一种基于 Transformer 的 DCAE-Transformer 模型,旨在改善健康状态(SOH)估计的准确性.该方法通过Pearson相关系数筛选关键特征,利用去噪自编码器(DAE)和卷积神经网络(CNN)相结合进行数据预处理和特征提取,再将数据输入 Transformer 框架完成预测.使用NASA和CALCE提供的数据集进行验证,DCAE-Transformer模型在 NASA电池样本上的误差指标(EMA、EMAP 和ERMS)均低于 1%,R2 值超过 99.5%;在 CALCE 样本上,误差指标低于 5%,R2 值超过 98%.结果表明,该模型在锂电池SOH 估计方面具有较高的精确性和泛化性.

Abstract

A novel model of DCAE-Transformer is introduced in this study,grounded in the Transformer architecture.It is designed to enhance the precision of state of health(SOH)estimation.Pearson correlation coefficient is employed for feature selection,denoising autoencoder(DAE)and convolutional neural network(CNN)amalgamation is integrated for data preprocessing and feature extraction.Subsequently,data is fed into the transformer framework for prediction.The validation conducted by using the datasets from NASA and CALCE demonstrates the metrics of exceptional performance for the DCAE-Transformer model:error indices(EMA,EMAP,and ERMS)on NASA battery samples remain below 1%,with the values of R-squared surpassing 99.5%;on CALCE samples,error metrics remain under 5%,with R-squared values exceeding 98%.These findings underscore the superior precision of the model and the generalizability in lithium battery SOH estimation.

关键词

锂电池/健康状态估计/卷积去噪自编码器/Transformer/预测性能

Key words

lithium battery/state of health estimation/convolutional denoising autoencoder(DCAE)/Transformer/predictive performance

引用本文复制引用

出版年

2025

三峡大学学报(自然科学版)

三峡大学

三峡大学学报(自然科学版)

CSTPCD北大核心

影响因子：0.401

ISSN：1672-948X

段落导航