EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Dual-Branch Prediction Framework for Lithium-Ion Battery SOC and SOH Based on Discharge-Derived Features and Convolutional Feature Analysis

Wanjun Wu, Tingjie Ba, Hongyi Zeng, Dong Jiao, Yongyuan Huang

IEEE Access · 2026 · 14 · 98417-98432

DOI: 10.1109/access.2026.3706717 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Wanjun Wu und 4 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Yunnan Power Grid Company.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate and interpretable estimation of the state of charge (SOC) and state of health (SOH) is critical to the safe and reliable operation of lithium-ion batteries. To address the issues of low accuracy and weak interpretability in conventional battery state prediction methods, this paper proposes a dual-branch SOC/SOH prediction framework based on stagewise discharge evolution characteristics and degradation-sensitive feature analysis. First, a unified four-dimensional input feature set is constructed by incorporating voltage, temperature, voltage derivative (dV/dt), and differential voltage (dV/dQ) to characterize both short-term discharge dynamics and long-term degradation behaviors. Then, a dual-branch deep learning framework is developed: the SOC branch uses a causal residual CNN-LSTM for real-time sequence tracking, and the SOH branch employs a CNN-BiLSTM-Attention structure to extract global degradation features for health regression. Tests on the Oxford battery aging dataset show the proposed method achieves favorable SOC prediction accuracy and competitive SOH estimation performance, with an SOC root-mean-square error (RMSE) of 0.53% and a mean absolute error (MAE) of 0.37%, and an SOH RMSE of 0.82% and an MAE of 0.65%, reaching a performance level comparable to support vector regression (SVR), which gives 0.81% RMSE and 0.67% MAE. Furthermore, convolutional feature response analysis reveals distinct activation patterns: SOC prediction relies on staged discharge characteristics and late-discharge voltage variations, while SOH estimation depends on dV/dt and dV/dQ derivatives related to voltage window shift, phase transition, and polarization increase. These results demonstrate the effectiveness and interpretability of the proposed dual-branch framework under standardized characterization-discharge conditions.

Förderung

  • Yunnan Power Grid Company

Zitieren

APA

Wu, W., Ba, T., Zeng, H., Jiao, D., & Huang, Y. (2026). A Dual-Branch Prediction Framework for Lithium-Ion Battery SOC and SOH Based on Discharge-Derived Features and Convolutional Feature Analysis. IEEE Access, 14, 98417-98432. https://doi.org/10.1109/access.2026.3706717

Vancouver

Wu W, Ba T, Zeng H, Jiao D, Huang Y. A Dual-Branch Prediction Framework for Lithium-Ion Battery SOC and SOH Based on Discharge-Derived Features and Convolutional Feature Analysis. IEEE Access. 2026;14:98417-98432. doi:10.1109/access.2026.3706717

BibTeX
@article{Wu2026,
  author = {Wu, Wanjun and Ba, Tingjie and Zeng, Hongyi and Jiao, Dong and Huang, Yongyuan},
  title = {A Dual-Branch Prediction Framework for Lithium-Ion Battery SOC and SOH Based on Discharge-Derived Features and Convolutional Feature Analysis},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {98417--98432},
  doi = {10.1109/access.2026.3706717}
}
Kostenlos starten →