State of Health Estimation of Lithium-Ion Batteries Based on Support Vector Regression with Bayesian Optimization

Accurate prediction of the state of health (SOH) of lithium-ion batteries (LIBs) is essential for ensuring their safe and efficient operation. In this study, we developed a Bayesian hyperparameter-optimized support vector regression (SVR) with a Radial Basis Function (RBF) model to analyze the experimental battery data and estimated SOH. The charging voltage difference of an equal time interval was selected as a feature to indicate the status trend of the battery. Pearson correlation coefficients were applied to confirm the relationship between the feature indicator and the practical SOH. The Lowess smoothing technique enhances the precision of the feature extraction and machine learning models by reducing noise and capturing the underlying trends in noisy datasets. The model's effectiveness was verified through various experiments, including small voltage differences. The proposed method was applied to the NASA datasets, yielding accurate SOH estimations based on the evaluation matrices of the model.