To address the challenge of rapid flow field prediction for supercritical airfoils, this paper proposes a hybrid deep learning model, termed TransCNN-FoilNet, based on two main approaches in current deep learning flow field prediction models: convolutional neural networks and Transformers. The model is capable of predicting the flow fields of supercritical airfoils with varying thicknesses at different angles of attack, achieving up to a 79.5% reduction in the mean absolute error compared to the baseline model. Additionally, the study introduces a new combined loss function for training the flow field prediction model, referred to as the weighted L1SSIM loss function. The results demonstrate that this loss function can improve the prediction of lift and drag coefficients, with the relative error in drag coefficient reduced by up to 17.8%. The proposed model achieves improved prediction accuracy and generalization performance while reducing complexity, providing a promising tool for fast and reliable flow field prediction of supercritical airfoils.