Real-world vehicle charging duration prediction based on transfer learning with SENet-CNN-Transformer Model

Charging duration, defined as the duration required to charge from current to target state of charge, critically determines user experience enhancement. Accurate charging duration prediction faces significant technical challenges, primarily stemming from two issues: the scarcity of real-world charging data and battery systems’ inherent deeply nested nonlinear features. To address these issues, we propose a charging duration prediction method based on the Squeeze-and-Excitation Network (SENet)-Convolutional Neural Network (CNN)-Transformer network. First, data-enhancement techniques are employed to improve the richness of the data set. Then, the SENet module is utilized to dynamically adjust feature weights through a channel attention mechanism. The CNN-Transformer model, combining the local feature extraction capabilities of CNN and the global modeling capacity of the Transformer, is employed to overcome the limitations of single networks in feature extraction and dependency modeling. The proposed model is pre-trained on laboratory data and then fine-tuned on real-world data using transfer learning techniques, significantly reducing model training time . Experiments on real-world battery data suggest that the SENet-CNN-Transformer model outperforms CNN-Transformer, Transformer, and Long Short-Term Memory models, with mean absolute error reductions of 56%, 65%, and 75%, respectively. With our transfer learning technique, the training time can be reduced by 4.5 times without sacrificing prediction accuracy. This approach boosts accuracy and cuts training time. A SENet-CNN-Transformer based charging duration prediction model is proposed. Transfer learning is adopted to integrate laboratory data and real-vehicle data. Data augmentation is used to reduce the impact of overfitting.