Space transformation search

Abstract

In this paper, a new evolutionary technique is proposed, namely space transformation search (STS), which transforms current search space to a new search space. By simultaneously evaluating solutions in current search space and transformed space, we can provide more chances to find solutions more closely to the global optimum and finally accelerate convergence speed. The proposed STS method can be applied to many evolutionary algorithms, and this paper only presents a STS based particle swarm optimization (PSO-STS). Experimental studies on 20 benchmark functions including 10 shifted functions show that the PSO-STS and its variations can not only achieve better results, but also obtain faster convergence speed than the standard PSO.