EA-based parameter tuning of multimodal optimization performance by means of different surrogate models

Abstract

In the current study, parameter tuning is performed for two evolutionary optimization techniques, Covariance Matrix Adaptation Evolution Strategy and Topological Species Conservation. They are applied for three multimodal benchmark functions with various properties and several outputs are considered. A data set with input parameters and metaheuristic outcomes is used for training four surrogate models. They are then each used by a genetic algorithm that is employed for searching the best parameter settings for the initial approaches. The genetic algorithm uses the model outputs as the direct fitness evaluation and only the best found parameter setting is tested within the original metaheuristics. Each model quality is priory evaluated, but they are all subsequently used in the search process to observe how the (in)accuracy influences the final result. Additionally, the genetic algorithm is used for tuning these approaches directly to test if the search conducts to the same parameter set, or at least close to it.