The surge in electric vehicle (EV) popularity necessitates innovative approaches for estimating the state of health (SOH) of EV lithium-ion batteries. This study introduces a transformer-based online SOH estimation model that leverages actual EV driving data, marking a departure from conventional methods that rely on lab-experimented battery cycle data. Our model comprises a transformer encoder and processes the raw sequences of battery voltage, current, state of charge, and vehicle speed. Despite the inherent noise in the EV battery readings while driving, the model shows high accuracy, with a mean absolute error of 1.31% and a root mean square error of 2.08%. Furthermore, this study unveils through self-attention map analysis that the model attends the stationary period of EVs to estimate the SOH. Although this study has a limitation in the dataset which lacks a wide range of driving route patterns, it still demonstrates the significant potential of transformer models in online SOH estimation for EVs while also providing valuable insights for future data collection.
Keywords Lithium-ion batteries, Electric vehicles, State of health estimation, Deep learning, Transformer