Exploiting Building Blocks in Hard Problems with Modified Compact Genetic Algorithm

In Evolutionary Computation, good substructures that are combined into good solutions are called building blocks. In this context, building blocks are common structure of high- quality solutions. The compact genetic algorithm is an extension of the genetic algorithm that replaces the latter’s population of chromosomes with a probability distribution from which candidate solutions can be generated. This paper describes an algorithm that exploits building blocks with compact genetic algorithm in order to solve difficult optimization problems under the assumption that we have already known building blocks. The main idea is to update the probability vectors as a group of bits that represents building blocks thus avoiding the disruption of the building blocks. Comparisons of the new algorithm with a conventional compact genetic algorithm on trap-function and traveling salesman problems indicate the utility of the proposed algorithm. It is most effective when the problem instants have common structures that can be identify as building blocks.