Multimodal Optimization Using GA in Specific Electromagnetic Field Problems

Abstract

This paper analyzes a multimodal optimization problem concerning the determination of the critical wavenumbers of a dielectric-conductor waveguide. The concomitant exploration around multiple optimal points is solved via genetic algorithms, by enabling both similitude and fitness-based replacements of the old solutions. Through the interaction between these components, the algorithm can balance between promoting different and precise solutions, thus becoming suitable for various objective landscapes with abrupt, large variations. As the genetic algorithm is mainly meant to preserve the diversity of the solutions, the best individuals depicted from the clusters of the final population are improved via an inertial gradient method and only the well-adapted ones are declared as results. The experimental investigations done for different configurations of the design problem demonstrate that this hybrid algorithm is able to provide convenient sets of optimal solutions.