A study of optimizing the performance of genetic algorithms using design-of-experiments in job-shop scheduling application

Genetic Algorithms (GAs) have received considerable amount of attention because of their potential as an optimization technique for complex combinatorial problems and have been successfully applied to scheduling problems. However, there is no particular guideline on how to select appropriate values for GA parameters. The overall objective of this paper is to develop an efficient design-of-experiment to study any possible effect for genetic algorithm (GA) factors, namely, the population size, number of generations, the rate of crossover, the rate of mutation, the length of the block, problem complexity, and more importantly their interaction effects on the solution quality. Job-shop scheduling problems (JSSPs) are used for this study. We performed two experiments; screening and sequential experiments. The results from both experiments do not match. If the experimentation range changes, different factors may become significant. Therefore, there is no guideline in setting GA parameters to obtain the best solution for JSSPs in this study.