Archive Update Strategy Influences Differential Evolution Performance

In this paper the effects of archive set update strategies on differential evolution algorithm performance are studied. The archive set is generated from inferior solutions, removed from the main population, as the search process proceeds. Next, the archived solutions participate in the search during mutation step, allowing better exploration properties to be achieved. The LSHADE-RSP algorithm is taken as baseline, and 4 new update rules are proposed, including replacing the worst solution, the first found worse solution, the tournament-selected solution and individually stored solution for every solution in the population. The experiments are performed on CEC 2020 single objective optimization benchmark functions. The results are compared using statistical tests. The comparison shows that changing the update strategy significantly improves the performance of LSHADE-RSP on high-dimensional problems. The deeper analysis of the reasons of efficiency improvement reveals that new archive update strategies lead to more successful usage of the archive set. The proposed algorithms and obtained results open new possibilities of archive usage in differential evolution.