Analysis of lithium-ion battery degradation prediction via machine learning

Accurate prediction of the remaining useful life of lithium-ion batteries is critical for enhancing safety, optimizing battery management systems, and improving operational efficiency in applications such as electric vehicles and renewable energy storage. Traditional machine learning models have been widely used for RUL estimation but often fail to effectively capture long-range dependencies and temporal variations in battery degradation. To address these limitations, this study proposes a transformer-based model that leverages self-attention mechanisms to model complex degradation patterns. Furthermore, we apply SHapley Additive exPlanations (SHAP) to assess feature importance. Experimental results demonstrate that the proposed model significantly outperforms traditional machine learning approaches. These findings highlight the potential of transformer architectures for improving the accuracy and reliability of lithium-ion battery health prediction, providing valuable insights for intelligent battery management systems.