Scholarlink

<p>The accelerating electrification of transport demands highly reliable lithium‑ion batteries (LIBs) with predictable aging behavior. However, the degradation of LIBs is governed by complex nonlinear interactions among material composition, chemistry, and operational stressors, creating significant uncertainty in capacity-loss prediction. This study aims to develop a mechanistically interpretable, optimization‑driven machine learning model that captures these interactions and enhances degradation forecasting accuracy. A curated dataset of 1221 data points, collected from published experimental studies, was standardized and analyzed to establish strong statistical and correlational foundations. Gradient Boosting Decision Tree (GBDT) models were separately optimized to assess their comparative performance. Cross‑validation within the training subset (90%) ensured the robustness of hyperparameter tuning, while a held‑out testing subset (10%) provided fair generalization evaluation. The optimized GBDT models demonstrated prediction accuracy exceeding 0.95 with consistently low RMSE values, outperforming conventional CNN, MLP, and SVR baselines. Explainability analysis via SHAP confirmed that cycle number and temperature dominate capacity degradation, while electrolyte composition and charge rate exert secondary but synergistic influences. The interpretability framework provides mechanistic insight linking electrochemical parameters to degradation dynamics, allowing data‑driven yet physically meaningful predictions. The developed workflow establishes a reproducible, optimization‑enhanced predictive tool suitable for both experimental researchers and battery management system developers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimization driven gradient boosting framework for robust prediction of lithium-ion battery degradation using explainable machine learning

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Anupam Yadav,
  • Mustafa Abdullah,
  • V. Vivek,
  • Ibrahim Khersan,
  • Nora Rashid Najem,
  • Prabhat Kumar Sahu,
  • Joshila Grace,
  • Vikas Wasoom,
  • Abdolali Yarahmadi Kandahari

摘要

The accelerating electrification of transport demands highly reliable lithium‑ion batteries (LIBs) with predictable aging behavior. However, the degradation of LIBs is governed by complex nonlinear interactions among material composition, chemistry, and operational stressors, creating significant uncertainty in capacity-loss prediction. This study aims to develop a mechanistically interpretable, optimization‑driven machine learning model that captures these interactions and enhances degradation forecasting accuracy. A curated dataset of 1221 data points, collected from published experimental studies, was standardized and analyzed to establish strong statistical and correlational foundations. Gradient Boosting Decision Tree (GBDT) models were separately optimized to assess their comparative performance. Cross‑validation within the training subset (90%) ensured the robustness of hyperparameter tuning, while a held‑out testing subset (10%) provided fair generalization evaluation. The optimized GBDT models demonstrated prediction accuracy exceeding 0.95 with consistently low RMSE values, outperforming conventional CNN, MLP, and SVR baselines. Explainability analysis via SHAP confirmed that cycle number and temperature dominate capacity degradation, while electrolyte composition and charge rate exert secondary but synergistic influences. The interpretability framework provides mechanistic insight linking electrochemical parameters to degradation dynamics, allowing data‑driven yet physically meaningful predictions. The developed workflow establishes a reproducible, optimization‑enhanced predictive tool suitable for both experimental researchers and battery management system developers.