A Vector Grouping Learning Brain Storm Optimization Algorithm for Global Optimization Problems

Abstract

The original brain storm optimization (BSO) method does not rationally compromise global exploration and local exploitation capability, which results in the premature convergence when solving complicated optimization problems such as the shifted or shifted rotated functions. To address this problem, this paper develops a vector grouping learning BSO (VGLBSO) method. In VGLBSO, the individuals’ creation based on a VGL scheme is first developed to improve the population diversity and compromise the global exploration and local exploitation capability. Moreover, a hybrid individuals’ update scheme is established by reasonably combing two different individuals’ update schemes, which further compromises the global exploration and local exploitation capability. Finally, the random grouping scheme, instead of K-means grouping, is allowed to shrink the computational cost and maintain the diversity of the information exchange between different individuals. Twenty-eight popular benchmark functions are used to compare VGLBSO with 12 BSO and nine swarm intelligence methods. Experimental results present that VGLBSO achieves the best overall performance, including the global search ability, convergence speed, and scalability among all the compared algorithms.