EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Towards Sustainable Battery Utilization: A Physics-Informed Multimodal Deep Learning Framework for Robust SOC Estimation

Lipeng Hu, Jinjun Tang, Li Fang

Sustainability · 30.10.2026 · 18 (21) · 11079

DOI: 10.3390/su182111079 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sustainability, online erschienen am 30.10.2026. Verfasst von Lipeng Hu und 2 weiteren Autor:innen. Die Arbeit nennt 58 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Furong Scholar Award Program of Hunan Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Achieving high-accuracy and robust State of Charge (SOC) estimation is critical not only for alleviating range anxiety but also for optimizing energy utilization and extending battery service life, which are fundamental to sustainable energy management in electric vehicles. However, existing approaches often suffer from limitations in estimation accuracy, physical consistency, and cross-scenario adaptability, which hinder their practical deployment and compromise the long-term operational sustainability of battery systems. To address these challenges, this study proposes a SOC estimation framework that integrates the strengths of physics-based modeling and data-driven learning, serving as an advanced sustainability monitoring tool. First, an improved SOC labeling strategy is implemented using a segment-level Coulomb counting-Extended Kalman Filter (SCC-EKF) algorithm. Second, a MultimodAl Transformer-based estimation model is developed, incorporating Physics-Informed constraints (PIDAT). By combining physical priors with the deep learning model, the framework effectively prevents SOC predictions that violate physical laws, thereby supporting safer and more efficient battery operation. Finally, evaluations on a real-world dataset comprising 106 electric buses demonstrate that the framework consistently outperforms state-of-the-art methods, achieving a 45.85% reduction in SMAPE volatility and over 35% relative error reduction in MAE. Specifically, in the Leave-One-Vehicle-Out (LOVO) cross-validation, it achieves an RMSE of 1.3112 in long-sequence scenarios (vs. 2.1251 for baselines), proving superior cross-vehicle adaptability. Furthermore, ablation studies indicate that the proposed architecture effectively prevents the non-linear amplification of errors, drastically reducing the Physics Constraint Violation Rate (PCVR) from 0.2550 to 0.0312. These experimental results prove that the proposed model can serve as a sustainable usage monitoring technology for power batteries, providing strong technical support for the sustainable development of green public transportation.

Förderung

  • Furong Scholar Award Program of Hunan Province

Zitieren

APA

Hu, L., Tang, J., & Fang, L. (2026). Towards Sustainable Battery Utilization: A Physics-Informed Multimodal Deep Learning Framework for Robust SOC Estimation. Sustainability, 18(21), 11079. https://doi.org/10.3390/su182111079

Vancouver

Hu L, Tang J, Fang L. Towards Sustainable Battery Utilization: A Physics-Informed Multimodal Deep Learning Framework for Robust SOC Estimation. Sustainability. 2026;18(21):11079. doi:10.3390/su182111079

BibTeX
@article{Hu2026,
  author = {Hu, Lipeng and Tang, Jinjun and Fang, Li},
  title = {Towards Sustainable Battery Utilization: A Physics-Informed Multimodal Deep Learning Framework for Robust SOC Estimation},
  journal = {Sustainability},
  year = {2026},
  volume = {18},
  number = {21},
  pages = {11079},
  doi = {10.3390/su182111079}
}
Kostenlos starten →