Hybrid Population Based MVMO for Solving CEC 2018 Test Bed of Single-Objective Problems

The MVMO algorithm (Mean-Variance Mapping Optimization) has two main features: i) normalized search range for each dimension (associated to each optimization variable); ii) use of a mapping function to generate a new value of a selected optimization variable based on the mean and variance derived from the best solutions achieved so far. The current version of MVMO offers several alternatives. The single parent-offspring version is designed for use in case the evaluation budget is small and the optimization task is not too challenging. The population based MVMO requires more function evaluations, but the results are usually better. Both variants of MVMO can be improved considerably if additionally separate local search algorithms are incorporated. In this case, MVMO is basically responsible for the initial global search. This paper presents the results of a study on the use of the hybrid version of MVMO, called MVMO-PH (population based, hybrid), to solve the IEEE-CEC 2018 test suite for single objective optimization with continuous (real-number) decision variables. Additionally, two new mapping functions representing the unique feature of MVMO are presented.