Evaluating the Effectiveness of Boosting and Bagging Ensemble Techniques in Forecasting Lithium-Ion Battery Useful Life

It is essential to forecast the exact rate at which the cell's capacity would decline for practical uses, to comprehend the intricate and non-linear behavior of the cell. Furthermore, the majority of studies provided subpar prediction criteria, making early cell lifetime prediction difficult. Applying reliable and accurate aging models to the dynamic on-road conditions presents additional challenges. In this work, the battery lifetime during its earliest phases of use was accurately predicted using machine learning models. After analyzing the patterns of the parameters, 12 hand-crafted features were selected and the raw data of the first 100 cycles of 126 cells was used for creating the dataset for the features. The dataset was then used to train five machine learning models namely random forest, gradient boosting machine (GBM), light gradient boosting machine (LGBM), extreme gradient boosting machine (XGBoost), and gradient boost with categorical features (CATBoost). The statistical analysis reveals that XGBoost algorithm present the best result with a R² value of 0.95 and root-mean-square-error (RMSE) of 97 cycles. Lastly, in comparison to existing studies, the RMSE significantly reduced from a maximum of 138 to 97 cycles.