Cooperative Velocity Updating model based Particle Swarm Optimization

Abstract

Particle swarm optimization (PSO) has shown its good performance in many optimization problems. This paper proposes a Cooperative Velocity Updating algorithm based Particle Swarm Optimization (CVUPSO), which is inspired by the competition and cooperation methods of different populations in natural swarm living, such as bees, ants, birds, fish, etc. In this algorithm, before an elite is introduced from other sub-swarms or a new particle is randomly born, the weak particle will be eliminated out of its sub-swarm. In each iteration process, every sub-swarm abandons a least potential particle. The CVUPSO recorded four special positions: pbest, lbest, gbest and lworst. The pbest represents the current particle’s best position while lbest represents the current sub swarm’s best position, and gbest is the best position among the whole swarm, lworst is the position of the particle with the worst performance. A new update method is adopted in CVUPSO, where the particles are more likely to follow lbest than follow gbest in the early stage of iteration, but opposite in the later stage. In this paper, two variants of CVUPSO are proposed, one variant is CVUPSO with Random inertia weight (for short CVUPSO-R), and another is CVUPSO with Exponential decline inertia weight (for short CVUPSO-E). By making comparative experiments on several widely used benchmark functions, analysis results show that the performance of these two improved variants are more promising than the recently developed PSO algorithms for searching multiple peak values of multiple objects optimization problem.