A TeLU-enhanced deep learning approach for state-of-health estimation of lithium-ion batteries

Hits:

Impact Factor:4.3

DOI number:10.1016/j.epsr.2026.113974

Journal:Electric Power Systems Research

Key Words:State of health; Lithium-ion batteries; Electric vehicles; Discrete wavelet transform; TeLU-enhanced

Abstract:Accurate estimation of the state of health (SOH) of lithium-ion batteries is critical for ensuring the safety and reliability of electric vehicles (EVs). However, existing methods often face challenges in real-world applications, particularly in accurate capacity calibration, robust feature extraction, and effective modeling of complex degradation patterns. To address these issues, this study proposes a comprehensive SOH estimation framework based on real-world operational data collected from EVs across diverse geographical regions. First, reliable SOH pseudo-labels are generated by coupling the reverse ampere-hour counting technique with the forgetting factor recursive least squares (FFRLS) algorithm to mitigate the interference of real-world data fluctuations. Second, both time-domain and time-frequency domain features are extracted, with the latter derived through the discrete wavelet transform (DWT) to fully leverage complementary degradation information. The core innovation of this paper lies in the comprehensive framework integration of dual-domain feature engineering and deep network design. Within this framework, the pioneering application of the Tangent Exponential Linear Unit (TeLU) activation function serves as a critical enabling component, which significantly improves the nonlinear representation capability and effectively resolves the gradient oscillation problems of conventional activations. Furthermore, a deterministic perturbation-based analysis method is introduced to quantitatively evaluate feature importance, thereby improving model interpretability. Experimental results demonstrate that the proposed method effectively handles real-world data challenges, achieving a MAE below 1.43% and a RMSE under 1.80%, demonstrating superior accuracy and stability compared to the implemented baseline models in strict cross-vehicle generalization tests.

Note:中科院3区

Co-author:Pengyu Liu,Lei Bao

First Author:Muyao Wu

Indexed by:Essay collection

Correspondence Author:Li Wang

Document Code:113974

Discipline:Engineering

Document Type:J

Volume:265

ISSN No.:0378-7796

Translation or Not:no

Date of Publication:2026-08-09

Included Journals:SCI、EI

Links to published journals:https://www.sciencedirect.com/science/article/pii/S0378779626012629