A State of Health Estimation Method for Lithium-ion Batteries Based on Duration of an Equal Discharging Voltage Difference and XGBoost

Accurate estimation of the state of health (SOH) of lithium-ion batteries is essential for their safe operation. With growing experimental data and advancements in machine learning, data-driven SOH estimation methods have gained prominence. While many studies focus on extracting health features from incremental capacity curves, noise from low sampling frequencies can hinder accuracy. In this paper, we propose a new health feature derived from the voltage curve: the Duration of an Equal Discharging Voltage Difference (DEDVD). This feature demonstrates high accuracy, is less affected by noise, and has lower requirements for sampling frequency. Based on this feature, we developed an XGBoost model that requires training on data from only a single battery to accurately capture degradation trends, achieving high accuracy in predictions for other batteries. The mean error (ME) across all batteries does not exceed 1.6%, and the root mean square error (RMSE) is under 0.7%. Furthermore, we compared four popular machine learning methods. Training on Battery 1 and estimating SOH for Batteries 2 to 7 showed that the XGBoost method outperformed the others overall.