Simulation-driven deep learning approach for module-level state of charge estimation in electric vehicles with active thermal management assessment

Abstract Accurate state of charge (SOC) estimation for electric vehicle (EV) lithium-ion batteries is challenged by complex electro-thermal dynamics. Despite recent advances in data-driven methods, a critical gap persists: existing models are trained on single-cell datasets and therefore fail to capture the cell-to-cell thermal gradients and dynamic electro-thermal coupling inherent in practical multi-cell modules. To address this gap, this study proposes a comprehensive, Simulation-based data-driven framework bridging high-fidelity simulations with a hybrid convolutional neural network-long short-term memory (CNN-LSTM) architecture. Following rigorous validation of the base pseudo-2D electrochemical-thermal model against experimental data, a 32-cell module dataset was generated under highly transient driving cycles (US06, UDDS), successfully bypassing the prohibitive costs of experimental data acquisition. Using the optimized 100-s historical data window, the CNN-LSTM achieved an MAE of approximately 1.1% on the held-out US06 dataset at an unseen temperature condition, demonstrating generalization under an operating condition excluded from model development. Five-fold cross-validation further yielded an average SOC MAE of approximately 2.0%, confirming the stability of the proposed model across different temporal subsets of the training data. Furthermore, an integrated 3D thermal management case study demonstrated that while an active spiral liquid cooling channel maintains optimal module temperatures (30–40 °C) and limits spatial gradients (≤ 5 °C), a cooling system failure leads to a rapid loss of usable capacity, reaching the voltage cut-off at approximately 1600 s, less than half the discharge duration sustained under active cooling. Ultimately, this research confirms that effective battery management requires both accurate deep learning algorithms for state estimation and active thermal control to maintain usable capacity and thermal safety.