A Hybrid Deep Learning Method for the Estimation of the State of Health of Lithium-Ion Batteries

Abstract

This paper proposes a method for estimating the state of health (SOH) of lithium-ion batteries (LIBs) using a combination of vision transformer (VIT) and gated recurrent unit (GRU) networks. The new scheme adopts a VIT to extract features from the battery measured data and incorporates a GRU network to mitigate the limitations of the VIT caused by positional encoding. The resulting VIT-GRU network is designed to comprehensively capture information relevant to the battery SOH. Simulation experiments on the NASA dataset illustrate the notable results achieved by the VIT-GRU, with prediction root mean square error (RMSE) and mean absolute error (MAE) up to 0.54% and 0.38%, respectively, demonstrating the exceptional performance of the VIT-GRU network in SOH estimation. Compared to other complex deep learning (DL) methods, the VIT-GRU significantly outperforms them, according to the RMSE and MAE of the predicted values.