Particle Swarm Optimization: Global Best or Local Best?

A number of empirical studies have compared the two extreme neighborhood topologies used in particle swarm optimization (PSO) algorithms, namely the star and the ring topologies. Based on these empirical studies, and also based on intuitive understanding of these neighborhood topologies, there is a faction within the PSO research community that advocates the use of the local best (lbest) PSO due to its better exploration abilities, diminished susceptibility to being trapped in local minima, and because it does not suffer from premature convergence as is the case with the global best (gbest) PSO. However, the opinions that emanated from these studies were based on a very limited benchmark suite containing only a few benchmark functions. This paper conducts a very elaborate empirical comparison of the gbest and lbest PSO algorithms on a benchmark suite of 60 boundary constrained minimization problems of varying complexities. The statistical analysis conducted shows that the general statements made about premature convergence, exploration ability, and even solution accuracy are not correct, and shows that neither of the two algorithms can be considered outright as the best, not even for specific problem classes.