Optimized XGBoost modeling for accurate battery capacity degradation prediction

Abstract

Lithium-ion batteries have notable benefits in their elevated energy power, density, and efficiency. However, the deterioration of capacity remains a prominent concern throughout their usage. Specifically, calculating the remaining capacity is essential for guaranteeing safe operations, which has prompted the creation of accurate capacity estimate models. Battery capacity estimation is one of the critical functions in the Battery Management System (BMS), and battery capacity indicates a battery's maximum storage capability, which is vital for the battery State of Charge (SOC) estimation and lifespan management. In order to increase the accuracy of battery capacity prediction, we provide in this work an improved Extreme Gradient Boosting (XGBoost) model that has been tuned by Random Search hyperparameter tweaking. Our proposed method achieved a R2 value of 0.9931%, a Mean Squared Error (MSE) of 0.0068, and a Root Mean Squared Error (RMSE) of 0.0825 when applied to an extensive dataset. These outcomes indicate significant improvements compared to traditional regression models. Moreover, the suggested technique has yielded superior accuracy and resilience in estimating lithium-ion battery capacity.

Keywords

• Battery capacity degradation
• Predictive modeling
• XGBoost optimization
• Hyperparameter tuning
• Random search optimization
• Energy storage systems