State-of-health prediction of Li-ion NMC Batteries Using Kalman Filter and Gaussian Process Regression

Abstract

State of health monitoring for batteries is of utmost importance for efficient and secured operations. This work proposes a hybrid approach to forecast battery’s performance losses. Particularly, the proposed method combines the Kalman filter (KF) and Gaussian Process Regression (GPR) techniques to predict the battery capacity evolution with aging. The effectiveness of the approach is validated based on experimental data. Data are obtained testing four cells of lithium nickel manganese cobalt oxide. These cells are cycled using a dynamic current profile derived from the Worldwide Harmonized Light Vehicles Test Cycle (WLTC) under controlled temperature conditions. The proposed method is validated by comparing the actual End of Life (EoL) with the predicted, one obtained with different sections of the training dataset; 30%, 50% and 70%. The results show that the best average prediction error is obtained when the training data set is larger, and the aging trend is uniform. The results also show that the dispersion around the estimated EoL is lower when the training data set is larger. For seven of the twelve case studies, the estimated EoL is lower than the actual one, which is a conservative but good scenario for safety reasons.