Scholarlink

<p>Accurate State of Charge (SoC) estimation is critical for reliability, safety, and longevity of lithium-ion batteries, especially in electric vehicles, energy storage systems, and portable electronics. While machine learning models such as extreme gradient boosting (XGBoost) offer promising performance for SoC estimation, their effectiveness is often limited by the quality of input features and the efficiency of hyperparameter tuning. This paper proposes a data-driven framework that integrates systematic feature engineering with a bio-inspired Driving Training-Based Optimization (DTBO) algorithm for hyperparameter tuning of an XGBoost regression model. The proposed approach is validated on two open-source battery datasets, achieving prediction accuracy with an RMSE of 0.0011, MAE of 0.0006, and R² of 0.9999. The results demonstrate that the combined XGBoost-DTBO framework significantly outperforms existing deep learning, hybrid, and physics-informed models, establishing a new benchmark for SoC estimation suitable for real-world battery management systems.</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

Accurate state of charge estimation of lithium-ion batteries using XGBoost optimized with driving training-based optimization

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Karthikeyan M,
  • Sugirtha T

摘要

Accurate State of Charge (SoC) estimation is critical for reliability, safety, and longevity of lithium-ion batteries, especially in electric vehicles, energy storage systems, and portable electronics. While machine learning models such as extreme gradient boosting (XGBoost) offer promising performance for SoC estimation, their effectiveness is often limited by the quality of input features and the efficiency of hyperparameter tuning. This paper proposes a data-driven framework that integrates systematic feature engineering with a bio-inspired Driving Training-Based Optimization (DTBO) algorithm for hyperparameter tuning of an XGBoost regression model. The proposed approach is validated on two open-source battery datasets, achieving prediction accuracy with an RMSE of 0.0011, MAE of 0.0006, and R² of 0.9999. The results demonstrate that the combined XGBoost-DTBO framework significantly outperforms existing deep learning, hybrid, and physics-informed models, establishing a new benchmark for SoC estimation suitable for real-world battery management systems.