Deep learning and data augmentation for robust battery state of charge estimation in electric vehicles

Accurate state of charge (SOC) estimation is critical for the effective management of lithium-ion batteries in electric vehicles (EVs). However, traditional SOC estimation techniques, which rely on a limited set of measurable parameters such as voltage-integrated time (VIT), are inadequate for capturing the complex dynamics of the battery across a range of operational conditions. This paper introduces a novel data-driven approach that enhances the conventional input features by incorporating both average power and average resistance. Leveraging these advanced features, the proposed method employs long short-term memory (LSTM) and recurrent neural network (RNN) architectures to achieve a more accurate and robust characterization of SOC. Compared to VIT-based models, our approach results in up to a 61% reduction in root-mean-square error (RMSE), with final RMSE values of 0.0083 for the LSTM model and 0.0154 for the RNN model. Furthermore, the mean absolute error decreases to 0.0106 for LSTM and 0.0120 for RNN. These findings demonstrate that the proposed methodology significantly improves generalizability and robustness across diverse operational conditions, offering a promising solution for data-driven battery management systems in EV applications.