Parallel computation models of particle swarm optimization implemented by multiple threads

Abstract

Particle Swarm Optimization (PSO) is an algorithm motivated by biological systems. However, PSO implementations are sequential, meaning that particles cannot simultaneously interact with other members in the same swarm. This study tries to develop an exact PSO model whose particles simultaneously interact with each other. To model limited communication capability, particles in a swarm are separated into several subgroups. Communication among the subgroups is implemented by parallel computation models based on broadcast, star, migration and diffusion network topologies. Due to the expense and difficulty of true parallel computation, multiple threads are used to model simultaneous particle interaction. We compare the four parallel PSO models and the traditional sequential computation model using measures of convergence error, generations to convergence and execution time. Three experiments to examine the performance of the parallel PSO models are also included.