On the critical battery electrochemical parameters across different phases of a single discharge process using a transformer framework
Journal
Energy Conversion and Management: X
Journal Volume
30
Start Page
101629
ISSN
25901745
Date Issued
2026-05
Author(s)
Abstract
Accurate characterization of electrochemical parameters is critical for interpreting the time- and frequency-domain responses of energy conversion devices and for enabling precise, efficient, and controllable management. In the context of lithium-ion batteries, this study introduces a deep learning (DL) framework using electrochemically derived simulation data to predict discharge curves and to distinguish the dominant internal electrochemical parameters for each of the three constant-current stages. Extracting attention weights from the transformer encoder helps identify the most influential parameters at each stage. Our study reveals subtle variations in the key electrochemical parameters that control discharge across different time-evolution stages. During the early stage of discharge, the behavior is mainly governed by positive-electrode parameters, such as the volume fraction of active material and the maximum concentration in the positive electrode. As the discharge progresses, however, negative-electrode parameters—particularly, the volume fraction of active material in the negative electrode, become increasingly influential. These outcomes are further verified through two additional operations: Sobol-based global sensitivity analysis and Shapley additive explanations. This DL framework reproduces the time-dependent battery behavior during a single discharge while elucidating the relationship between electrochemical parameters and battery response, thereby enabling efficient parameter assessment or identification and rational voltage-window selection for battery applications.
Subjects
Attention weight matrix
Electrochemical parameters
Lithium-ion battery
Sensitivity analysis
SHAP analysis
Publisher
Elsevier Ltd
Type
journal article