Abstract:

Traditional antenna optimization designs need numerous simulation trials of different parameter combinations to reach the optimum, which leads to low efficiency in solving high dimensional antenna design and optimization problems. To address this issue, we design an initial Kriging model by using a few uniformly distributed sampling points and their simulation data. During the optimization iterations, the population of each generation is comprised of individuals with high fitness as well as individuals with high diversity. The optimal individual is selected according to its responses and uncertainty predicted by the Kriging model. Electromagnetic simulations are conducted for this individual, and the results are used to update the Kriging model. This algorithm is applied to optimize the resonant frequencies of an E-shaped antenna with 6 variables. Compared with other optimization methods, the number of EM simulation is reduced by about 80%.

Key words: antenna design, high dimensional optimization, Kriging, uniformly sampling