A demonstration of the utility of fractional experimental design for finding optimal genetic algorithm parameter settings

This paper demonstrates that the use of sparse experimental design in the development of the structure for genetic algorithms, and hence other computer programs, is a particularly effective and efficient strategy. Despite widespread knowledge of the existence of these systematic experimental plans, they have seen limited application in the investigation of advanced computer programs. This paper attempts to address this missed opportunity and encourage others to take advantage of the power of these plans. Using data generated from a full factorial experimental design, involving 27 experimental runs that was used to assess the optimum operating settings of the parameters of a special genetic algorithm (GA), we show that similar results could have been obtained using as few as nine runs. The GA was used to find minimum cost schedules for a complex component assembly operation with many sub-processes.