An evolution strategy and genetic algorithm hybrid: An initial implementation and first results

Abstract

Evolution Strategies (ESs)[15] and Genetic Algorithms (GAs)[13] have both been used to optimise functions, using the natural process of evolution as inspiration for their search mechanisms. The ES uses gene mutation as it's main search operator whilst the GA mainly relies upon gene recombination. This paper describes how the addition of a second mutation operator, used in conjunction with the mutation and crossover operators of the normal GA, can improve the GA's performance on rugged fitness landscapes. We then show that by adding Lamarckian replacement the GA's performance on smooth landscapes can also be improved, further improving it's performance on rugged landscapes. We explain how the extra operators allow the GA to gain and exploit local information about the fitness landscape, and how this local random hill climbing can be seen to combine the search characteristics of the ES with those of the GA.