Back

Petkovic A.  

Global search PSO-like algorithm RC1

A novel, low-population PSO-like algorithm for solving global optimization problems is proposed. Performance of subject algorithm is tested on the set of 12 benchmark functions under standard and non-standard (dimensionality increase up to 30000, domain extension up to 5 times and goal functions rotation) test conditions. For 10 of the considered problems, astonishing results are observed – convergence speed of the “universal” and several problem-dependent configurations of RC1 avail with quasi-invariant behavior in regard of problem dimensionality. Moreover, for 9 out of the 10 problems (standard test co nditions) referent configurations reach the solution in just 11-25 function evaluations, which corresponds to on-going progress of high-population PSOs in the initialization stage or the first iteration. Favorable results are also obtained for the cases of domain extension and test functions rotation. RC1 faces certain difficulties on the remaining 2 problems.


To reports list