Scholarlink

<p>Battery state-of-health (SoH) estimation is inherently challenging, requiring consideration of complex degradation processes and diverse battery characteristics. Complex degradation mechanisms and variability in battery behaviour must be accounted along with diverse battery characteristics. This research aims to develop a precise SoH estimation model. Such a model would guide optimal charging strategies and enhance understanding of battery behaviour over time. The methodology employs a stacking technique for mapping different types of dependencies. Linear, nonlinear, and long-temporal relationships are all addressed through this approach. Base learners include linear regression (LR), extreme gradient boosting (XGBoost), and random forest (RF). Each algorithm was selected to address specific battery characteristics. A novel preprocessing approach called cycle-preserved data expansion (CPDE) was developed. This technique maintains cycle count while ensuring uniform cycle length in the augmented data. Hyperparameters were optimised using grid search cross-validation (GridSearchCV). This involved an exhaustive search over specified parameter sets to enhance model robustness. The model was tested on both real-time and NASA datasets. Results showed significant improvements in evaluation metrics compared to conventional approaches. The findings confirm that combining multiple algorithms through stacking successfully capture degradation mechanisms. This results in more precise SoH predictions, enabling smarter battery management and contributing to longer battery lifespans.</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

Enhancing Battery Health Monitoring Using a Stacking Approach for Precise and Real-Time State-of-Health Estimation

  • 期刊论文

  • 发表日期：

    2025年7月25日

查看全文
  • Riya Sharma,
  • Anju Bala,
  • Ashima Singh,
  • Mukesh Singh

摘要

Battery state-of-health (SoH) estimation is inherently challenging, requiring consideration of complex degradation processes and diverse battery characteristics. Complex degradation mechanisms and variability in battery behaviour must be accounted along with diverse battery characteristics. This research aims to develop a precise SoH estimation model. Such a model would guide optimal charging strategies and enhance understanding of battery behaviour over time. The methodology employs a stacking technique for mapping different types of dependencies. Linear, nonlinear, and long-temporal relationships are all addressed through this approach. Base learners include linear regression (LR), extreme gradient boosting (XGBoost), and random forest (RF). Each algorithm was selected to address specific battery characteristics. A novel preprocessing approach called cycle-preserved data expansion (CPDE) was developed. This technique maintains cycle count while ensuring uniform cycle length in the augmented data. Hyperparameters were optimised using grid search cross-validation (GridSearchCV). This involved an exhaustive search over specified parameter sets to enhance model robustness. The model was tested on both real-time and NASA datasets. Results showed significant improvements in evaluation metrics compared to conventional approaches. The findings confirm that combining multiple algorithms through stacking successfully capture degradation mechanisms. This results in more precise SoH predictions, enabling smarter battery management and contributing to longer battery lifespans.