Investigation of number of children, number of parents, tournament size, and elitism in genetic algorithms

Abstract

The Genetic Algorithm (GA) can provide a means to solve many difficult optimization problems, notably problems with multi-modal and/or discontinuous functions. The GA can also address combinatorial (mixed discrete and continuous design variable) optimization problems. However, in providing these capabilities, the GA generally requires a large number of function evaluations. Many variations of the GA have been posed to reduce the computational cost and improve the solutions obtained. Among the intriguing options are: varying the number of designs considered in the tournament selection operator, varying the number of parents and children used in the crossover operator, and varying the number of elite individuals passed from one generation to the next. To examine the effects of these variations for a wide range of problem types, an investigation was made using several different problems, including multi-modal, highly constrained, and combinatorial problems. The results appear to be somewhat problem dependent, but a few general observations and conclusions can be made about these variations.