EUVIMEDEUVIMED entdecken
Crossref

Explainable Machine Learning Algorithms for State-of-Health Estimation and Remaining Useful Life Prediction of Lithium-Ion Batteries

Rajanand Patnaik Narasipuram, Aymen Flah, Alireza Hosseinpour

Journal of Modern Technology · 22.11.2026 · 494-503

DOI: 10.71426/jmt.v3.i2.pp494-503 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Modern Technology, online erschienen am 22.11.2026. Verfasst von Rajanand Patnaik Narasipuram und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

State-of-health (SOH) estimation and remaining useful life (RUL) prediction address related but distinct aspects of lithium-ion battery degradation. This study evaluates separate XGBoost regressors for SOH estimation and RUL prediction using four NASA Ames lithium-ion cells with complete capacity-fade trajectories. The reference capacity is defined as the median capacity of the first five discharges, while end of life (EOL) is defined as the first subsequent cycle at which capacity falls to or below 80% of this reference. At a fixed observation point of 600~s, the prediction features comprise electrical, thermal, cycle-related, initial-reference, and preceding electrochemical impedance spectroscopy (EIS) measurements. Full-discharge capacity and future EOL information are used only to construct retrospective target labels and are excluded from the corresponding predictors. A leave-one-cell-out evaluation yields 351 evaluated cycles. XGBoost achieves pooled SOH and RUL mean absolute errors (MAEs) of 1.157 percentage points and 17.948 cycles, respectively, while ridge regression achieves 1.195 percentage points and 17.611 cycles. The held-out B0006 cell exhibits a substantially higher XGBoost RUL MAE of 34.886 cycles, highlighting the limitations of cross-cell prognostic transfer despite low pooled SOH error. TreeSHAP analysis identifies the change in 600-s voltage relative to the cell's initial reference as the largest average contribution to SOH predictions. Removing this feature increases SOH root-mean-square error (RMSE) from 1.473 to 2.929 percentage points, while reducing RUL RMSE from 22.731 to 20.836 cycles. These findings demonstrate that feature importance and predictive utility differ between SOH estimation and RUL prediction.

Zitieren

APA

Narasipuram, R. P., Flah, A., & Hosseinpour, A. (2026). Explainable Machine Learning Algorithms for State-of-Health Estimation and Remaining Useful Life Prediction of Lithium-Ion Batteries. Journal of Modern Technology, 494-503. https://doi.org/10.71426/jmt.v3.i2.pp494-503

Vancouver

Narasipuram RP, Flah A, Hosseinpour A. Explainable Machine Learning Algorithms for State-of-Health Estimation and Remaining Useful Life Prediction of Lithium-Ion Batteries. Journal of Modern Technology. 2026:494-503. doi:10.71426/jmt.v3.i2.pp494-503

BibTeX
@article{Narasipuram2026,
  author = {Narasipuram, Rajanand Patnaik and Flah, Aymen and Hosseinpour, Alireza},
  title = {Explainable Machine Learning Algorithms for State-of-Health Estimation and Remaining Useful Life Prediction of Lithium-Ion Batteries},
  journal = {Journal of Modern Technology},
  year = {2026},
  pages = {494--503},
  doi = {10.71426/jmt.v3.i2.pp494-503}
}
Kostenlos starten →