Machine Learning-based Prediction of the Remaining Life of Sodium-ion Batteries

Sodium-ion battery capacity degradation exhibits nonlinearity and complexity, making accurate prediction of remaining useful life (RUL) extremely difficult. Therefore, this paper proposes an RUL prediction model combining Singular Spectrum Analysis (SSA) optimization algorithm with Gradient Boosting Regression Tree (GBRT). Firstly, experimental data is filtered and smoothed, and Incremental Capacity (IC) curves are plotted. The health indicators (HI) with high correlation to capacity decay are extracted from the IC curves. Secondly, PCA algorithm is used to reduce data redundancy, and the processed data is inputted into the GBRT model. SSA algorithm is used to search for the optimal hyperparameters to improve prediction accuracy. Finally, multiple sets of aging experimental data are used for validation, with RMSE, MAPE, and MAE all below 15.2, 7%, and 11.2, respectively. The results show that the proposed model has high prediction accuracy and robustness and outperforms other mainstream algorithms.