Machine Learning‐Based Real‐Time State of Health Estimation for Electric Vehicle Batteries in VANET Using GRU Networks and On‐Board Diagnostic Data

The recent growth of electric vehicles (EVs) and vehicular ad‐hoc networks indicates the need for accurate and lightweight health state estimation in EV batteries. Lithium‐ion batteries are the major contributors in this field. Real‐time health state estimation is important to confirm the performance and lifetime of EVs, particularly in a VANET‐enabled environment where battery status must be continuously shared across connected vehicles. They share information across connected vehicles. Traditional physical‐based estimations often struggle with high computational demands and are not suitable for complex practical scenarios. To address this problem in the upcoming state of health (SOH) estimation, we propose an adaptive, lightweight attention‐augmented gated recurrent unit (AL‐GRU) model. The GRU model captures the temporal dependencies in battery operational data by using gated recurrent units that selectively update information at each time step, and enable the model to effectively study degradation patterns. Then, an attention mechanism is applied to focus only on the important time steps. The suggested model restricts its branches to avoid unnecessary computational complexity. The lightweight design of the proposed model makes it suitable for VANET‐connected vehicles. The model is implemented in PyTorch and demonstrates strong predictive performance in terms of mean absolute error (MAE) of 0.039, root mean square error (RMSE) of 0.092, and R 2 of 0.859. Residual analysis further confirms the model's suitability for a real‐time VANET environment for practical battery management applications.