Enhanced life cycle prediction of lithium-ion batteries using tuned XGBoost models

Accurate prediction of the Remaining Useful Life (RUL) of lithium-ion batteries (LIBs) is critical for electric vehicles and energy storage systems, as battery health impacts performance, safety, and cost. Traditional methods often fail to capture the complex, nonlinear degradation processes of LIBs, which are influenced by factors such as temperature, current, and internal resistance. Machine learning (ML) models offer a compelling alternative by learning patterns directly from large datasets without relying on detailed physical models. This study explores data-driven ML approaches to predict battery discharge capacity and cycle life using early-cycle data, identifying key indicators of degradation. A comparison of ML algorithms, including SVM, KNN, neural networks, LightGBM, XGBoost, and linear regression, found that XGBoost was the most accurate, effectively capturing nonlinear degradation trends. Hyperparameter tuning using Randomized Search optimized parameters such as learning rate and tree depth, resulting in a Mean Absolute Percentage Error (MAPE) of 5.8% and improving predicted cycle counts by 48 cycles compared to the baseline. It demonstrates the potential of ML models to address the complexity of LIB degradation and provide scalable, accurate RUL predictions.