Scalability of a heterogeneous particle swarm optimizer

Most particle swarm optimization (PSO) algorithms maintain swarms of homogeneous particle, where all of the particles in the swarm follow the same behavior as specified via the particle position and velocity update rules. Many different position and velocity update rules have been developed, exhibiting different exploration - exploitation finger prints. Recently, a heterogeneous PSO (HPSO) has been developed which allows particles to randomly select a different behavior at each iteration from a behavior pool. At any time, the swarm consists of particles following different search behaviors. It was shown in [1] that the HPSO significanly outperformed a selection of homogeneous PSO algorithms on a set of classical benchmark functions. This article conducts an analysis of the scalability of the HPSO to large dimensional instances of the benchmark functions, in comparison with homogeneous PSO algorithms. It is shown that the HPSO is significantly more scalable than the homogeneous PSO algorithms used in this study.