State of Health Estimation for Lithium-Ion Batteries Using Multiple Indirect Feature Extraction and Bayesian Optimization–Transformer Model

Accurate estimation of the state of health (SOH) of lithium-ion batteries is a critical prerequisite for ensuring their safe and stable operation. However, the complex operating environments in practical applications often result in raw data containing significant noise and outliers, posing substantial challenges to SOH estimation. To this end, this paper proposes a SOH estimation method based on indirect health feature extraction with Bayesian optimization–transformer model. First, multiple indirect health features that reveal deeper battery health information are extracted to effectively mitigate the interference from noise and poor-quality data. Second, the Bayesian optimization algorithm is employed to tune the hyperparameters of the transformer model. Finally, the effectiveness of the proposed method is validated on three large publicly available data sets. Experimental results demonstrate that the proposed method achieves a mean absolute error and root-mean squared error for SOH estimation within 1.5% and 2%, respectively, which demonstrated its high accuracy and reliability.