Robust state of health estimation of commercial lithium-ion batteries based on enhanced hybrid machine learning model for electrified transportation

Lithium-ion batteries (LIBs) are widely used in electric vehicles and grid energy storage systems due to their high energy and power density, durability, and low cost. The accurate estimation of battery states mainly the state of health (SOH), to ensure the system stability, is a challenging task. Significant disparities arises in estimating the SOH using the conventional models for various chemistries of battery. These disparities stem from limited diversity in model selection, generalization, and insufficient hyperparameters tuning. To overcome these issues, five machine learning-based models have been discussed in this paper. The hybrid combinations and predictive framework with a systematic approach to estimate SOH for different types of batteries are also developed. To improve model generalization ability and hyperparameters tuning, leave-one-out cross-validation and grid search method are used in this paper. The detailed model generalizations and comparisons were verified by conducting experimental aging tests on four different types of battery chemistries in our laboratory and using publicly available NASA spacecraft battery datasets. The models precisely estimate SOH, achieving estimation errors MAE, MSE, and RMSE less than 0.01% and R² nearly 1, on our laboratory tested and NASA datasets. Whereas proposed models show 73% and 72% computational efficiency on both collected datasets, respectively, demonstrating the superior performance of the proposed models. Furthermore, its generalization ability can be utilized across various battery types.