High-Dimensional Multi-Objective Optimization Using Co-operative Vector-Evaluated Particle Swarm Optimization with Random Variable Grouping

Vector-evaluated particle swarm optimization (VEPSO) is a particle swarm optimization (PSO) variant which employs multiple swarms to solve multi-objective optimization problems (MOPs). Each swarm optimizes a single objective and information is passed between swarms using a knowledge transfer strategy (KTS). The recently proposed co-operative VEPSO (CVEPSO) algorithm has been shown to improve the performance of VEPSO by decomposing the search space into subspaces of lower dimensionality. However, the effectiveness of CVEPSO is heavily dependent on the strategy used to group variables together, because isolating dependent variables leads to performance degradation. This paper explores the use of a random grouping technique within CVEPSO to increase the probability of allocating interacting variables to the same subcomponent. The results demonstrate that random grouping significantly improves performance of the CVEPSO algorithm, especially in high-dimensional environments. Additionally, CVEPSO with random grouping is shown to perform competitively with other top multi-objective optimization algorithms.