Abstract: In the design and optimization of structural cross-sections, the efficient and accurate evaluation of aerodynamic performance is of significant importance. To address the inefficiency in iterative computations involving diverse design parameters‌, this study proposed an intelligent prediction method based on a deep learning surrogate model, focusing on the rapid and accurate prediction of flow fields around two-dimensional combined bluff body cross-sections and static force coefficients under steady wind conditions. Specifically, this method employed a unified image-like shape representation to characterize the aerodynamic shapes of bluff body combined cross-sections, ensuring broad applicability without being limited by specific cross-section configurations. By integrating convolutional attention mechanisms and residual modules to construct the neural network architecture and using mean squared error to capture prediction errors, the method achieved a highly nonlinear mapping from aerodynamic shapes to flow characteristics and static force coefficients. The model achieves prediction errors within ‌3.7% for velocity fields‌, ‌0.35% for surface pressure‌, and ‌6.25% for force coefficients‌ under steady wind, meeting all accuracy requirements. Additionally, the computational efficiency was improved by four orders of magnitude compared to conventional CFD. This method provides an efficient and practical solution for the rapid prediction of aerodynamic performance for bluff body cross-sections under steady wind environments.