Updating Strategy in Compact Genetic Algorithm Using Moving Average Approach

Abstract

The compact genetic algorithm (cGA) has a distinct characteristic that it requires almost minimal memory to store candidate solutions. It represents a population structure as a probability distribution over the set of solutions. Although cGA offers many advantages, it has a limitation that hinges on an assumption of the independency between each individual bit. For example, cGA fails to solve a deceptive function or the so called trap function, which is a standard difficult test problem for genetic algorithm. This paper proposes applying a moving average technique to update a probability vector in the compact genetic algorithm. This method requires fewer evaluations and achieves a higher solution quality. The results are compared with the original cGA, sGA, persistent elitist cGA (pe-cGA) and nonpersistent elitist cGA (ne-cGA). The compared results illustrate that the proposed methodology can successfully improve the solution quality by modifying the updating strategy of cGA