Hind El Ouazzani

The rapid growth in lithium‐ion battery applications has increased the demand for efficient and accurate models capable of predicting battery behavior. Electrochemical models provide high fidelity but are typically too computationally intensive for real‐time control or large‐scale simulations. Herein, a deep learning‐based surrogate model (DL‐SBM) designed to forecast battery performance using historical data, significantly reducing computational costs while maintaining accuracy is introduced. The surrogate models are trained on synthetic datasets generated by an electrochemical simulator under diverse charging scenarios, and optimized through adaptive hyperparameter tuning. The resulting models accurately predict critical battery parameters such as voltage, temperature, and state of charge. Furthermore, a reinforcement learning case study focused on extreme fast charging demonstrates that the DL‐SBM achieves accuracy comparable to physics‐based simulators, while operating ≈225 times faster, requiring less memory, and ensuring robust performance under extreme conditions. These results underline the suitability of the DL‐SBM for real‐world battery management applications.