State of health (SoH) estimation is crucial for the reliable operation of battery management systems. While various SoH estimation approaches have been proposed, the integration of adaptive filtering with practical indirect health measurements remains insufficiently explored. This study introduces an online, data-driven SoH estimation framework that combines Gaussian process regression (GPR) with an extended Kalman filter (EKF). Moreover, the proposed method uses the normalized maximum incremental capacity measurement as a health indicator (HI). Extracting this HI during the constant current phase of the constant current constant voltage charging protocol enables online SoH estimation. The equivalent full cycle count is used for a priori prediction, while the HI is employed for a posteriori updates. Experimental validation of the method is carried out using three publicly available battery datasets, i.e., Dataset 1, Dataset 2, and Dataset 3, through a leave-one-battery-out cross-validation under varying operational conditions. The proposed GPR-EKF outperforms the EKF on Datasets 1-2 and is comparable on Dataset 3, while outperforming the Long Short-Term Memory-EKF across all datasets with average root mean square errors (RMSEs) of 1.88%, 0.45%, and 1.06%, respectively. Furthermore, the method exhibits robust SoH estimation, maintaining an RMSE of 0.91% even when the HI measurement is intermittently available with an 80% probability. These results highlight the potential of the proposed GPR-EKF method for accurate, robust, and online SoH estimation in practice.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)