Differential Evolution Optimization Algorithm for Electromagnetic Device Design with High-dimensional Mixed Discrete-Continuous Variables

Optimization of electromagnetic device design with mixed discrete and continuous variables is a challenge, especially for those with high-dimensional discrete variables. An orthogonal adaptive discrete variable selection based differential evolution (OADVSDE) is proposed to solve such problems in this paper. Continuous variables use the canonical differential evolution algorithm to perform the genetic operations, while the discrete variables are processed mainly in two steps. The first step is conducted by using orthogonal experimental design to sample a small number of representative combinations. Then good combinations are given more chances to generate more promising offspring in the following generations. To verify its effectiveness, this algorithm is first tested on multi-objective benchmark suites. A design problem of five layers stacked dielectric resonator antenna which involves eight discrete variables with seven possible values for each discrete variable, is also considered. The relevant results show that OADVSDE is superior to the compared algorithm. It can be found that the proposed method is feasible for solving optimization problems with high-dimensional mixed discrete and continuous variables.