Battery State of Charge Prediction Based on Conformer Architecture

With the rapid development of electric vehicles (EVs), advances in battery management technology have led to significant improvements in battery safety and cycle life, However, accurate state of charge (SOC) estimation still faces many challenges, such as overcharging, over-discharging and temperature variations. Traditional SOC estimation methods have limitations like cumulative errors and high computational complexity, while deep learning techniques provide new ideas to solve this problem. In this paper, we propose a Conformer-based SOC estimation model that integrates Transformer and Convolutional Neural Network (CNN) architectures. It leverages a multi-head self-attention mechanism to capture global dependencies and convolutional modules to extract local temporal features, combined with sliding window techniques for dynamic data modeling. Experimental results demonstrate that the Conformer reduces the Mean Absolute Error (MAE) and Mean Squared Error (MSE) by approximately 95% compared to traditional long short-term memory (LSTM) and CNN models, showcasing its superior accuracy, robustness, and generalization ability. It highlight the potential of the Conformer model for advancing SOC estimation in EVs.