Accurate estimation of state of charge and state of health of electric vehicle power batteries based on dynamic channel pruning informer-stacked denoising autoencoder model

Accurate estimation of state of charge (SOC) and state of health (SOH) of electric vehicle (EV) power batteries is critical for users to plan their trips and improve EV safety. The informer model, an improved variant of the transformer, excels in long-sequence time-series forecasting tasks; however,...

Full description

Saved in:
Bibliographic Details
Published in:Electrical engineering Vol. 108; no. 1; p. 19
Main Authors: Peng, Ziran, Wang, Shunhao, Zhang, Shengchun, Meng, Shuangwu
Format: Journal Article
Language:English
Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.01.2026
Springer Nature B.V
Subjects:
Accuracy
Aging
Attention
Batteries
Cleaning
Economics and Management
Efficiency
Electric charge
Electric vehicles
Electrical Engineering
Electrical Machines and Networks
Electrolytes
Energy Policy
Engineering
Estimates
Estimation
Inference
Lithium
Machine learning
Methods
Neural networks
Noise reduction
Original Paper
Power Electronics
Pruning
Real time
Root-mean-square errors
Sensors
Stability
State of charge
Transformers
Dynamical channel pruning
Enhanced informer
Stacked denoising autoencoder
State of health
Electric vehicle
State of charge
Power batteries
ISSN:0948-7921, 1432-0487
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Accurate estimation of state of charge (SOC) and state of health (SOH) of electric vehicle (EV) power batteries is critical for users to plan their trips and improve EV safety. The informer model, an improved variant of the transformer, excels in long-sequence time-series forecasting tasks; however, the number of heads in its multihead probsparse attention and multihead self-attention usually needs to be set manually, and redundant heads not only lead to overparameterization but also reduce estimation accuracy. This paper introduces the dynamic channel pruning (DCP) technique to highlight dominant attention features, which enhance the performance and stability of the informer model. Many estimation methods overlook the necessity of data cleaning; therefore, this paper proposes stacked denoising autoencoder (SDAE) to clean voltage, current, and temperature data to provide higher quality data for the informer model. Finally, the cleaned data are input into the dynamic channel pruning informer (DCPInformer) model to estimate SOC and SOH. Experimental validation is conducted using battery datasets from the University of Oxford and Beijing Institute of Technology. Results indicate that the mean absolute error (MAE) and root mean square error (RMSE) of the DCPInformer-SDAE model reach 0.25% and 0.38%, respectively, for estimating SOC, and 0.51% and 0.64%, respectively, for estimating SOH. In the inference stage, DCPInformer-SDAE exhibits higher efficiency; compared with traditional transformer, long short-term memory, gated recurrent unit, and extreme learning machine models, it achieves lower inference latency and higher throughput, while also improving accuracy, stability, and generalizability in SOC and SOH estimation.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0948-7921
1432-0487
DOI:10.1007/s00202-025-03351-w