EUVIMEDEUVIMED entdecken
Crossref

Long-Term Prediction for EV Battery Aging Trend: A Deep Learning Approach

Zihui Lin, Hao Ouyang, Yuntao Zou, Zhichun Liu, Dagang Li

IEEE Transactions on Transportation Electrification · 2026-08 · 12 (4) · 6032-6041

DOI: 10.1109/tte.2026.3686259 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Transportation Electrification, 2026-08. Verfasst von Zihui Lin und 4 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Gefördert von: National Key Research and Development Program of China, National Natural Science Foundation of China, Key-Area Research and Development Program of Hubei Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Electric vehicles (EVs) have become pivotal in the automotive industry’s shift toward clean energy. However, battery aging significantly impacts charge storage capacity, affecting driving range, performance, and operational costs. Accurate prediction of battery aging trends is crucial for effective battery management, yet remains challenging due to the complex interplay of factors such as driving habits, charging patterns, environmental conditions, and battery state. This study proposes a novel framework to predict EV charge electricity within the 40%–90% state-of-charge (SOC) interval, leveraging its direct correlation with battery health. Using real Internet-of-Vehicle (IoV) data from 15 ride-hailing EVs in Guangzhou, collected over one year at 20-s intervals, multidimensional features—including vehicle operation, charging habits, environmental conditions, and battery status—are extracted. A deep learning model architecture is introduced, which employs multiscale embedding and patch-based Transformer encoder to capture temporal patterns across different features while efficiently processing long-term dependencies through patch-level attention mechanisms. The multiscale embedding hierarchically extracts features at short, medium, and long temporal granularities, while the patch-based encoder operates on segmented time-series representations to enable efficient multivariate time-series forecasting of battery aging trends. The model is trained on 14 vehicles’ full-year data, fine-tuned on one target vehicle’s first half-year data using transfer learning, and tested via rolling forward prediction for the second half-year. Comparative experiments demonstrate superior performance, with overall mse, MAE, RMSE, MAPE, and MSPE values of 5.94, 1.72, 2.44, 0.031, and 0.0018, respectively.

Förderung

  • National Key Research and Development Program of China
  • National Natural Science Foundation of China
  • Key-Area Research and Development Program of Hubei Province
  • Project of Characteristic Innovation in General Colleges and Universities of Guangdong Province
  • Key Scientific Research Project of Higher Education Institutions in Guangdong Province
  • Fund for the Development of Science and Technology

Zitieren

APA

Lin, Z., Ouyang, H., Zou, Y., Liu, Z., & Li, D. (2026). Long-Term Prediction for EV Battery Aging Trend: A Deep Learning Approach. IEEE Transactions on Transportation Electrification, 12(4), 6032-6041. https://doi.org/10.1109/tte.2026.3686259

Vancouver

Lin Z, Ouyang H, Zou Y, Liu Z, Li D. Long-Term Prediction for EV Battery Aging Trend: A Deep Learning Approach. IEEE Transactions on Transportation Electrification. 2026;12(4):6032-6041. doi:10.1109/tte.2026.3686259

BibTeX
@article{Lin2026,
  author = {Lin, Zihui and Ouyang, Hao and Zou, Yuntao and Liu, Zhichun and Li, Dagang},
  title = {Long-Term Prediction for EV Battery Aging Trend: A Deep Learning Approach},
  journal = {IEEE Transactions on Transportation Electrification},
  year = {2026},
  volume = {12},
  number = {4},
  pages = {6032--6041},
  doi = {10.1109/tte.2026.3686259}
}
Kostenlos starten →