Differential evolution with large initial populations

Abstract

This paper proposed a novel method to determine which individuals can enter from the first search phase to the second phase search. An orthogonal array constructs the initial population. The first search phase is neighborhood-based search, and game theory is also introduced. After finishing the first phase, there are two criteria to enter the next phase. One is a traditional standard, fitness. Another is the score, which is generated from the game. This new algorithm, named OGLSHADE-CS, involves other techniques: linear population reduction, success history base adaption, multi-strategy mutation, and conservative selection. This algorithm and some state-of-the-art algorithms test the 2020 CEC benchmark suite. They are compared using some statistic tests. The results show that game theory can improve performance.