Ensemble learning prediction model for lithium-ion battery remaining useful life based on embedded feature selection

To prevent potential dangers, it is crucial to accurately predict the Remaining Useful Life (RUL) of Lithium-Ion Batteries (LIBs) as capacity gradually decreases during use. Based on the Stanford MIT Battery Life Data, our researchers find the selection of input features difficult. Therefore, a new embedded Feature Selection (FS) method (LS-Embed-BHRSA) was proposed, which combines the Laplace score (LS) algorithm with the Binary Hybrid Reptile Search Algorithm (BHRSA). It is an organic fusion method of filtering and wrapping, composing an embedding structure that can quickly and accurately find the best feature subset. In addition, a Stacking multimodel integration strategy was also proposed to completely use the advantages of different models and improve the prediction accuracy of LIBs' RUL. The validation results showed that the Hybrid Reptile Search Algorithm (HRSA) performed the best on the 2022 test function, and LS-Embed-BHRSA found the best features in comparison with other FS architectures on the UCI classification data. The testing results of the MIT data show that the overall proposed prediction model exceeds other machine learning models in metrics, such as RMSE, MAE, MAPE and R2, indicating its strong competitiveness.