A Data-Based Approach for Aging Estimation of Railway Vehicles Propulsion Batteries
IEEE Access · 2026 · 14 · 60984-60996
DOI: 10.1109/access.2026.3685321 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Do-Hun Kwon und 3 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Ministry of Climate, Energy, Environment (MCEE) and Korea Institute of Energy Technology Evaluation and Planning.
Abstract (Originalsprache)
The risk of fires and malfunctions in batteries supplying propulsion power to railway vehicles increases with age, making accurate aging estimation a crucial task for safe operation. However, conventional aging estimation methods face significant limitations: experimental methods require complex battery removal processes unsuitable for short maintenance cycles, and model-based approaches involve substantial computational costs for large-scale battery systems with high voltage and large capacity. These operational constraints indicate the need for intelligent data-driven approaches that can leverage periodically collected operational data without physical battery testing. However, conventional data-driven methods experience accuracy deterioration due to insufficient training data in the early commercialization stage of eco-friendly railway vehicles. This paper proposes a Combined Learning (CL) methodology that integrates transfer learning and continual learning to address these challenges. The proposed three-stage algorithm begins with pre-training using abundant EV battery data to establish initial feature extraction capabilities, followed by initial model adaptation with limited railway vehicle data, and continuous model updates leveraging periodic maintenance cycles. A time-series-aware rehearsal mechanism prevents catastrophic forgetting during continual learning. The appropriate AI model for the proposed CL method was selected through comparison, and battery degradation was measured from operational data including voltage, current, and temperature. Experimental validation demonstrates that the proposed CL method achieves Root Mean Square Error (RMSE) of 2.21% and Mean Absolute Error (MAE) of 1.77%, representing approximately 13% error reduction compared to conventional learning methods.
Förderung
- Ministry of Climate, Energy, Environment (MCEE) and Korea Institute of Energy Technology Evaluation and Planning
Zitieren
APA
Kwon, D., Na, S., Kim, B., & Cho, I. (2026). A Data-Based Approach for Aging Estimation of Railway Vehicles Propulsion Batteries. IEEE Access, 14, 60984-60996. https://doi.org/10.1109/access.2026.3685321
Vancouver
Kwon D, Na S, Kim B, Cho I. A Data-Based Approach for Aging Estimation of Railway Vehicles Propulsion Batteries. IEEE Access. 2026;14:60984-60996. doi:10.1109/access.2026.3685321
BibTeX
@article{Kwon2026,
author = {Kwon, Do-Hun and Na, Seok-Jin and Kim, Beom-Jun and Cho, In-Ho},
title = {A Data-Based Approach for Aging Estimation of Railway Vehicles Propulsion Batteries},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {60984--60996},
doi = {10.1109/access.2026.3685321}
}
EUVIMED