ABSTRACT
This chapter presents descriptions of some of the principal variants under the Particle Swarm Optimization (PSO) metaheuristic. In common with all evolutionary computation and Swarm Intelligence algorithms, the PSO metaheuristic consists of a population of agents that move iteratively in the search space of an optimization problem. The essence of the velocity update rules in global-best PSO is that a particle explores the search space randomly but constantly feels an attractive force towards the best location, it has found so far and the best location found by the swarm so far. The local-best variant of PSO, also called lbest PSO, involves a single change to the dynamical equations of gbest PSO. Lbest PSO generally takes more iterations to converge. This entails a larger number of fitness evaluations, which increases the overall computational cost of the method since the cost of fitness evaluation in regression problems always dominates over that of the PSO dynamical equations and bookkeeping.
