Abstract: Evolution algorithm (EA) is widely used for the optimization of various electromagnetic (EM) structures, however, its efficiency is generally low because of the random search in parameter space and numerous simulation trials. To address this problem, an efficient EM structure optimization algorithm which combines constrained differential evolution (DE) with Kriging model is proposed in this paper. According to reference designs, a tube space is first established by the algorithm, the area of evolution is restricted in this tube space by parameter transformation. By learning the samples in the tube and their simulation data, Kriging model replaces the EM solver to predict the responses of each individual after evolution. Comparing with the entire parameter space, the area of DE searching and Kriging learning is significantly reduced, therefore, the optimization efficiency is promoted. The proposed algorithm is validated by the optimization of a directional waveguide coupler, and it outperforms other existing algorithms in the quality of the solution and the convergence rate.