A fair comparison between standard PSO versions

Too often, when comparing a set of optimization algorithms, little effort, if any at all, is spent for finding the parameter settings which let them perform at their best on a given optimization task. Within this context, automatizing the choice of their parameter settings can be seen as a way to perform fair comparisons between optimization algorithms. In this paper we first compare the performances of two standard PSO versions using the “standard” parameters suggested in the literature. Then, we automatically tune the parameter values of both algorithms using a meta-optimization environment, to allow the two versions to perform at their best. As expected, results obtained by the optimized version are substantially better than those obtained with the standard settings. Moreover, they generalize well on other functions, allowing one to draw interesting conclusions regarding the PSO parameter settings that are commonly used in the literature.