Efficient surrogate-based antenna design optimization using novel sampling methods

Commonly used optimization methods require hundreds or even thousands of evaluations of objective function. Due to the high computational cost of electromagnetic (EM) simulation, the efficiency of directly using these methods is often very low. This paper presents an efficient surrogate-based antenna design optimization method in which the number of simulations is decreased by a novel sampling strategy. The key ideas are: (1) Limited initial sampling points are uniformly distributed in the parameter space by an improved Latin hypercube sampling algorithm. (2) In each optimization loop, a new sampling point chosen from several candidates is used to update the surrogate according to its uniformity and estimated fitness value. The operation and performance of the proposed method are demonstrated using two antenna design examples. Comparing with genetic optimization, the number of EM simulation is reduced by 75% and 79% respectively.