Dynamic Vector-Evaluated PSO with Guaranteed Convergence in the Sub-Swarms

Optimisation problems with more than one objective, of which at least at least one changes over time and at least two are in conflict with one another, are referred to as dynamic multi-objective optimisation problems (DMOOPs). The dynamic vector evaluated particle swarm optimisation (DVEPSO) algorithm is a co-operative particle swarm optimisation (PSO)-based algorithm and each of its sub-swarms solves only one objective function. The sub-swarms then share knowledge with one another through the particles´ velocity update. The default DVEPSO algorithm uses global best (gbest) PSOs as its sub-swarms. The guaranteed convergence PSO (GCPSO) algorithm prevents stagnation by forcing the global best particle to search within a defined region for a better solution. Using GCPSO results in proven convergence to at least a local optimum. Therefore, it is guaranteed that DVEPSO will converge to at least a local Pareto-optimal front (POF). This study investigates the effect of using GCPSOs as sub-swarms of DVEPSO. The results indicate that the GCPSO version of DVEPSO outperforms the gbest PSO DVEPSO on type I DMOOPs and in slowly changing environments.