Topological optimization and surrogate models are two mainstream methods in structural optimization. In large structures with complex load conditions, directly applying these methods for optimization can lead to low solution efficiency due to model complexity. This paper introduces a hybrid method that integrates sub-modeling techniques with topological optimization and surrogate modeling to enhance computational efficiency in optimization tasks. The proposed method has been employed in the optimization of the underframe of a railway freight car body. Initially, sub-models of the underframe beam structures were established and topologically optimized using the bi-directional evolutionary structural optimization (BESO) method. Subsequently, the Kriging model was employed for light weighting the underframe structure. Results show that the mass of the beam structures was reduced by 9.7% and the mass of the car body by 7.8%. The method demonstrated a significant reduction in computational efficiency by compared to calculations performed on the entire model, which can offer guidance for the structural optimization of complex structures.