Influence of the Archive Size on the Performance of the Dynamic Vector Evaluated Particle Swarm Optimisation Algorithm Solving Dynamic Multi-Objective Optimisation Problems


Many real-world problems consist of multiple objectives that are in conflict with one another and dynamic in nature. These kinds of problems do not have a single solution, but a set of optimal trade-off solutions. These trade-off solutions are stored in a fixed-size archive during the optimisation process. A decision maker then decides which trade-off solution to use for a specific optimisation problem. Larger archives require more computations than smaller archives. However, no research has been conducted to determine the influence of the archive size on the performance of algorithms when solving these kinds of problems. Therefore, this paper investigates the effect of archive sizes on the performance of the dynamic vector evaluated particle swarm optimisation algorithm. In addition, this study investigates the effect of the sampling size of the true Pareto-optimal front (POF) on the performance measure values when using small archives. The results indicate that the archive size does influence the performance of the algorithm and that in certain cases a small archive size may be beneficial. Furthermore, the results indicate that a larger sampling size of the true POF results in a worse performance measure value for the smaller archives.