A stochastic adaptive genetic algorithm for solving unconstrained multimodal numerical problems

In this paper, we investigate an adaptive genetic algorithm which will be able to identify the best combination of crossover and mutation operators in execution time. The adaptation involves four crossover methods (simple, arithmetical, non-uniform arithmetical and linear) and three mutation mechanism (uniform, non-uniform and creep). We validate the algorithm using some multimodal benchmarks function well known in the literature. Furthermore, using the ANOVA method and the Tukey test we proved that, in general, the adaptive algorithm works better than the static choice of the operators. Results show that even though some operators dominate the other ones, the use of other operators in the earlier stages of the algorithm can affect the quality of the solutions positively. Moreover, the use of an adaptive algorithm tends to evolve solutions faster than the other ones.