Hybrid State-of-Charge Estimation Method for Lithium-Ion Batteries Considering Uncertainty Quantification and Innovation Adaptation
Journal of The Electrochemical Society · 28.05.2026 · 173 (10) · 100516
DOI: 10.1149/1945-7111/ae6e9b ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of The Electrochemical Society, online erschienen am 27.05.2026. Verfasst von Zhenghong Wu und 4 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Gefördert von: Basic Research Program of Jiangsu Province, Changzhou University College Student Innovation and Entrepreneurship Training Program.
Abstract (Originalsprache)
Accurate estimation of the State of Charge (SOC) for lithium-ion batteries faces dual challenges of complex internal nonlinear characteristics and external noise interference. To address the lack of confidence evaluation in existing single data-driven methods and the limitations of the traditional Unscented Kalman Filter (UKF) in adapting to dynamic noise, this paper proposes a dual-adaptive GRU-UKF hybrid estimation framework that combines Monte Carlo Dropout (MC-Dropout) for uncertainty quantification and Innovation Adaptive Estimation (IAE) for noise adaptation. First, a second-order RC equivalent circuit model is established, and a Gated Recurrent Unit (GRU) neural network integrating the MC-Dropout mechanism is constructed. This network not only achieves end-to-end SOC prediction but also quantifies the epistemic uncertainty through probabilistic inference, thereby enabling dynamic adjustment of the confidence weight assigned to the data-driven observations within the UKF. Second, the IAE mechanism is introduced to monitor the innovation sequence of voltage residuals in real time and online correct the voltage measurement noise covariance matrix, which effectively compensates for the system bias caused by the parameter drift of the physical model. Finally, the proposed method is validated under various dynamic test profiles, including DST, BJDST, and US06, at an ambient temperature of 25 °C using the public CALCE dataset from the University of Maryland. Experimental results indicate that the hybrid method exhibits excellent accuracy and robustness across all test conditions. Under the DST profile with severe current fluctuations, the proposed method Root Mean Square Error (RMSE) is 2.06% and Mean Absolute Error (MAE) is 1.52%. Particularly under the highly random BJDST and the high-rate US06 profiles, the RMSE remains stable at 1.96% and 2.21%, with MAEs of 1.48% and 1.68%, respectively. Compared to single data-driven models and the standard UKF algorithm, the proposed hybrid method significantly reduces estimation errors and effectively resolves the estimation divergence issue in highly dynamic environments.
Förderung
- Basic Research Program of Jiangsu Province
- Changzhou University College Student Innovation and Entrepreneurship Training Program
Zitieren
APA
Wu, Z., Xu, W., Jiang, M., Sun, C., & Qiang, H. (2026). Hybrid State-of-Charge Estimation Method for Lithium-Ion Batteries Considering Uncertainty Quantification and Innovation Adaptation. Journal of The Electrochemical Society, 173(10), 100516. https://doi.org/10.1149/1945-7111/ae6e9b
Vancouver
Wu Z, Xu W, Jiang M, Sun C, Qiang H. Hybrid State-of-Charge Estimation Method for Lithium-Ion Batteries Considering Uncertainty Quantification and Innovation Adaptation. Journal of The Electrochemical Society. 2026;173(10):100516. doi:10.1149/1945-7111/ae6e9b
BibTeX
@article{Wu2026,
author = {Wu, Zhenghong and Xu, Weiyue and Jiang, Mingyu and Sun, Chengxiang and Qiang, Hao},
title = {Hybrid State-of-Charge Estimation Method for Lithium-Ion Batteries Considering Uncertainty Quantification and Innovation Adaptation},
journal = {Journal of The Electrochemical Society},
year = {2026},
volume = {173},
number = {10},
pages = {100516},
doi = {10.1149/1945-7111/ae6e9b}
}
EUVIMED