State of charge estimation for lithium-ion batteries base on convolutional neural network and transformer framework enhanced by the nutcracker optimization algorithm

Accurate estimation of the state of charge of lithium-ion batteries is essential for ensuring the safety and efficiency of battery management systems. Traditional prediction methods face significant challenges in handling the complexity and adaptability of battery data. With the rapid advancement of deep learning techniques in data modeling, this study proposes a novel convolutional neural network-Transformer-based deep learning model for state of charge prediction. Specifically, convolutional neural networks are employed to extract intricate features from raw battery data, while the Transformer architecture enhances feature representation and prediction accuracy through its global dependency modeling and multi-head attention mechanism. To further improve the model's performance and stability, the Nutcracker Optimization Algorithm is introduced to optimize key hyperparameters of the neural network. Experiments conducted under four temperature conditions demonstrate that the proposed method achieves high prediction accuracy across varying operational environments, with overall errors maintained below 0.03. During the discharge period of 40 %–100 % state of charge, estimation errors remain around 0.01. Moreover, comparative evaluations against commonly used neural network models, using five performance metrics, confirm the proposed model's superior robustness and accuracy in state of charge estimation.