State of health estimation for lithium-ion batteries with optimized DRT features

Accurate state of health (SOH) estimation is critical for enhancing battery safety and operational reliability. The distribution of relaxation times (DRT) provides valuable insights of electrochemical impedance spectroscopy (EIS) on the timescales. However, DRT results are highly reliant on parameter selection. Inappropriate parameters may lead to inaccurate SOH estimation results. Motivated by this, this paper proposes a novel method to optimize the DRT features. The preprocessing method is introduced to address the feature inconsistency. The particle swarm optimization (PSO) is employed to jointly optimize multiple DRT curve parameters during the aging process, strengthening the Pearson correlation coefficient between features and SOH. Verification confirms the effectiveness of the proposed feature optimization method. The results show that the root mean squared error (RMSE) decreased by more than 35.2%, R² increased by more than 14.0%, and mean absolute percentage error (MAPE) decreased by more than 21.0%. Accurate SOH estimation results can be achieved through the extreme gradient boosting (XGBoost), support vector regression (SVR), and Gaussian process regression (GPR) models.