A Bayesian Optimized Transformer Neural Network Model for State of Health Estimation of Lithium-Ion Batteries

The estimation of the state-of-health (SOH) of lithium-ion batteries based on data is a critical area of research in battery management systems. This paper suggests a SOH prediction method that integrates a Transformer neural network and Bayesian optimization to enhance the accuracy and stability of SOH prediction for lithium-ion batteries. Initially, the raw battery cycle data are cleansed, normalized, and structurally transformed. The input features are then constructed into a time series tensor that is tailored to the Transformer model. To address the challenge of optimizing the key hyperparameters in the Transformer network, the Bayesian optimization framework is implemented to facilitate automatic parameter tuning and enhance the model’s accuracy. The experimental validation of the proposed method is conducted using three sets of data samples, B5, B7, and B18, from the NASA Battery Public Dataset. The evaluation indexes are considerably superior to the comparative prediction models, and the method exhibits excellent generalization performance and fitting ability on all types of battery degradation curves. The technique described in this paper is capable of accurately predicting SOH trends. It is also scalable and adaptable, making it an effective data-driven solution for the assessment of battery remaining life and condition management.