Graded Particle Swarm Optimization (GPSO)

Particle Swarm Optimization has attracted a vast amount of researchers over the past two decades for its ability to search global optima by simple update equations for position and velocity. A set of randomly initialized points in a search space exchange information regarding their position and fitness. The individual's personal best factor and the group leader's positions support the swarm's movement towards optima. However, this strategy is not guaranteed to achieve the global optimum values in all the scenarios, which is otherwise dependent on the complexity of the fitness function. In addressing this issue, a better performing variant is proposed by introducing the concept of grouping, gradation (hierarchy) and universal leader concepts. In the proposed variant, during initialization, the members of the swarm are grouped for search and each group is associated with a group leader based on best fitness individual in the group. All these group leaders form a hierarchical group associated with a universal leader. For, applying this strategy a fourth term in the PSO velocity update equation is introduced governing the influence of a universal leader. In the current paper, the GPSO algorithm, its modified equation and deduction of the general PSO rules followed by its evaluation on the standard benchmark functions are presented. This algorithm like other swarming techniques is expected to support new search, exploration and mapping strategies in the fields of computational intelligence, swarm intelligence and swarm robotics.