Work

State-of-health estimation of lithium-ion batteries using end-of-charge features with cross-domain validation

Year: 2026

Type: article

Abstract: Conventional lithium-ion battery state-of-health (SOH) estimation methods often rely on complete charge–discharge cycles or are restricted to specific battery chemistries and protocols, hindering their generalization across diverse real-world scenarios. This study proposes a robust SOH estimation framework utilizing an end-of-charge (EOC) fingerprint window. By establishing a unified voltage–current zone independent of dataset-specific distributions, the method extracts statistical features from high-voltage responses and low-current decay patterns near charge completion. To simultaneously capture intra-cycle local variations and cross-cycle degradation trends, a hybrid architecture termed CCT-Net (CNN–CBAM–Transformer) is developed. Within this framework, a Convolutional Neural Network (CNN) identifies local patterns, a Convolutional Block Attention Module (CBAM) adaptively enhances critical features, and a Transformer models long-range degradation dependencies. Furthermore, SHAP analysis is integrated to quantify feature contributions and interpret their evolution across various degradation stages. Experimental results on the XJTU dataset demonstrate superior accuracy, with an average Mean Absolute Error (MAE) of 0.45% across six operating conditions. Extensive validation across 15 cross-domain scenarios—encompassing four datasets, two chemistries (NCM and LFP), capacities ranging from 1.1 to 20 Ah, and diverse protocols—yields an average MAE of 1.11% using only a single target-domain battery for adaptation. These results demonstrate that EOC fingerprint features effectively capture degradation dynamics under compound domain shifts, providing a compact and robust solution for SOH estimation using partial charging data in practical battery management systems.

Source: Journal of Energy Storage

DOI: 10.1016/j.est.2026.124183

Authors: Fuyuan Liang, Xing Zhou, Haiqing Yu, Tao Zhang, Yajie Liu

Institution: National University of Defense Technology

Language: English

Topic: Advanced Battery Technologies Research

Subfield: Automotive Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Affordable and clean energy

Open Access status: hybrid

Estimated APC paid: $3,560