Exclusion radius particle swarm optimizer

Abstract

Function optimization is a problem that has existed since modern engineering began. With multiple variables, brute-forcing or simple regression models become unfeasible. This paper proposes a shrinking population control method for a dynamic population particle swarm optimizer. The proposed control mechanism creates an exclusion zone, where particles are periodically purged from the simulation. The selection of particles purged relies on their particle best i.e. the position where it found its personal best fitness. Particles with their particle best outside of the exclusion zone with a predetermined radius from the current global best are removed from the simulation. The purging is done periodically in stages of equal iteration counts, and the radius is shrunk by a factor after every stage. By testing 5 benchmark mathematical functions, the results show that the proposed population control mechanism achieves better final optima than only a basic particle swarm optimizer and a simple shrinking dynamic population particle swarm optimizer where the worst performing particle is removed every so often.