On-the-fly crossover adaptation of genetic algorithms

Genetic algorithms (GAs) are widely used to solve complex optimization problems, and many variations of GAs have been proposed including several kinds of crossover operations. However, there have been few works, in which more than one crossover operators were used in a GA implementation. This paper presents an adaptive strategy, which selects a crossover operator to be used not in advance but dynamically during the algorithm execution. To select an appropriate crossover operator among given two kinds of crossover operators, for each pair of chromosomes (individuals) to be crossed over, we propose a new measure called the elite degree. The elite degree shows the potential proficiency of an individual in a specific generation. Experimental results for benchmark test functions show the effectiveness of the proposed method with the adaptive crossover selection based on the elite degree