Extract features from lithium-ion battery electrochemical impedance spectra and estimate state of health based on improved convolutional autoencoder-temporal convolutional network

Predicting the state of health (SOH) of batteries is crucial for understanding their remaining lifespan and formulating more effective maintenance and management strategies. Utilizing electrochemical impedance spectroscopy (EIS) to assess the health status of batteries offers advantages such as high...

Full description

Saved in:
Bibliographic Details
Published in:Ionics Vol. 31; no. 5; pp. 4261 - 4279
Main Authors: Lou, Cheng, Wang, Shi, Li, Zhaoting, Wang, Kai
Format: Journal Article
Language:English
Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.05.2025
Springer Nature B.V
Subjects:
Aging
Alternative energy
Artificial neural networks
Batteries
Chemistry
Chemistry and Materials Science
Condensed Matter Physics
Deep learning
Electric vehicles
Electrochemical impedance spectroscopy
Electrochemistry
Energy Storage
Feature extraction
Life span
Lithium-ion batteries
Machine learning
Neural networks
Optical and Electronic Materials
Real time
Renewable and Green Energy
Spectrum analysis
Time dependence
Temporal convolutional network
State of health
Unsupervised feature extraction
Electrochemical impedance spectroscopy
ISSN:0947-7047, 1862-0760
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Predicting the state of health (SOH) of batteries is crucial for understanding their remaining lifespan and formulating more effective maintenance and management strategies. Utilizing electrochemical impedance spectroscopy (EIS) to assess the health status of batteries offers advantages such as high precision, rapid response, non-invasiveness, and reliability for accurately forecasting the remaining lifespan of batteries. In this paper, we propose an innovative method combining electrochemical impedance spectroscopy data to efficiently recognize and process complex patterns using deep neural networks by converting EIS into a two-dimensional image format. We have developed an improved convolutional autoencoder (ICAE) optimized to extract key features directly related to battery capacity in 2D EIS images and significantly improve feature characterization. The optimized features are further fed into the temporal convolutional network (TCN) to perform SOH prediction tasks. TCN utilizes its powerful time-dependent capture capability and long sequence memory mechanism to demonstrate superior performance in the field of SOH estimation. Compared with traditional methods, the proposed strategy not only significantly increases the prediction accuracy, but also opens up a new way to understand and analyze the internal relationship between complex time series and image data.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0947-7047
1862-0760
DOI:10.1007/s11581-025-06226-z