Increasing Genetic Algorithm Efficiency for Wire Antenna Design Using Clustering

Abstract

The Genetic Algorithm (GA) is a very robust, powerful technique that is capable of optimizing designs in a very multimodal search spaces. However, it also requires significant numbers of simulations to perform such optimizations. If the simulations are expensive, as in the case of antenna design, GA's can be prohibitively expensive to use. A clustering technique has been investigated which cuts the required number of function calls 20-90% with minor or no degredation in the optimization quality. In this technique, a GA using real-valued genes is halted when the population has clustered around portions of the search space, and a local optimization technique completes the optimization quickly. This method has been applied to a variety of test functions and wire antenna designs, and the advantages of this technique seem to have broad applicability.