A comparison of evolutionary algorithms on a set of antenna design benchmarks

Many antenna design and optimization problems require optimizing multimodal, high dimensional, non-convex and inseparable objective functions. This has led researchers towards stochastic optimization techniques such as evolutionary algorithms (EAs) instead of classical gradient-based methods for these applications. However, despite many past successes, very little is known about which types of EAs map best to which types of antenna optimization problems. The goal of this work is to investigate this mapping of EAs to applications by comparing the performance of three EAs on five benchmark antenna design problems and one real-world problem derived from a NASA satellite mission. Performance of these algorithms has been compared on the basis of success rates and average convergence time over 30 independent runs. Our results indicate that age-layered population structure genetic algorithm (ALPS-GA) performed best in terms of success rates and convergence speed. However, on the NASA antenna design problem differential evolution achieved highest success rates, which was marginally better than ALPSGA. We also explored the effect of increasing antenna complexity on the antenna gain.