IEEE Transactions on Transportation Electrification (Early Access), DOI: https://doi.org/10.1109/TTE.2026.3714009

A Fusion-Based SOH Estimation for Lithium-ion Batteries with SHAP-Guided Feature Extraction and Generative Data Augmentation

Dandan Liu¹, Guanwei Jiang¹, Simin Peng¹, Daohan Zhang², Yuanliang Wang¹, Yan Ma², and Michael Pecht<sup>3

1</sup>School of Electrical Engineering, Yancheng Institute of Technology, Yancheng, China
²Department of Control Science and Engineering, Jilin University, Changchun, China
³Center for Advanced Life Cycle Engineering (CALCE), University of Maryland, College Park, MD, USA

For more information about this article and related research, please contact Prof. Michael Pecht.

Abstract:

Accurate state of health (SOH) estimation of lithium-ion battery is crucial for ensuring the safety and reliability of battery systems. Existing data-driven SOH estimation methods suffer from a heavy reliance on manual feature extraction, which often fails to effectively capture complex battery degradation characteristics and capacity regeneration phenomena, particularly in small-sample scenarios. This limitation consequently restricts the adaptability and generalization capability of these methods. To address these challenges, a SOH estimation framework that integrates deep feature extraction with generative adversarial networks is developed. The approach combines iTransformer with bidirectional long short-term memory networks, as a means to automatically extract health features from temporal battery state data. A Shapley additive explanation method is then employed to evaluate the contribution of each health feature, allowing for the identification of the most significant features for SOH estimation. The proposed method is validated using NASA and Oxford battery datasets containing multiple full life-cycle degradation processes. Assessment of this approach against numerous experimental results demonstrate that the developed approach achieves high SOH estimation accuracy, with errors consistently remaining below 0.7%, while maintaining acceptable computational efficiency for practical battery management system applications.

This article is available online here and to CALCE Consortium Members for personal review.