Solving dynamic multi-objective problems with vector evaluated particle swarm optimisation

Abstract

Many optimisation problems are multi-objective and change dynamically. Many methods use a weighted average approach to the multiple objectives. This paper introduces the usage of the vector evaluated particle swarm optimiser (VEPSO) to solve dynamic multi-objective optimisation problems. Every objective is solved by one swarm and the swarms share knowledge amongst each other about the objective that it is solving. Not much work has been done on using this approach in dynamic environments. This paper discusses this approach as well as the effect of the population size and the response methods to a detected change on the performance of the algorithm. The results showed that more non-dominated solutions, as well as more uniformly distributed solutions, are found when all swarms are re-initialised when a change is detected, instead of only the swarm(s) optimising the specific objective function(s) that has changed. Furthermore, an increase in population size results in a higher number of non-dominated solutions found, but can lead to solutions that are less uniformly distributed.