Diversity-based selection pooling scheme in evolution strategies

Actual implementations of standard Evolution Strategies (ES) suffer from premature convergence. This results from greediness in the ranking-based selection scheme as well as a lack of feedback on the efficacy of mutation step adaptation. This paper proposes a simple modification to minimize adverse effects of the above-mentioned evolution operators on algorithm performance by a diversity-based pooling scheme. Tests on several benchmark problems vindicate the virtue of this modification. The modified algorithm also sheds light on convergence status of the population with respect to a given optimization task, hence allowing further improvements toward premature convergence checking and parallel search algorithms.