Genetic algorithms and an indifference-zone ranking and selection procedure under common random numbers for simulation optimization

Simulation modeling is a widely used descriptive tool for modeling stochastic systems. However, when the goal is optimization, simulation modeling must be used in conjunction with an optimization algorithm. Genetic algorithms (GA) is one of many optimization methodologies that have been used for optimization of simulation models. The general idea with GA is that through iterations a population of alternative solutions is able to adapt to a given environment by means of the stochastic processes of selection, recombination, and mutation.; The most critical step with GA is the assignment of the selective probabilities to the alternatives. Selective probabilities are the GA guidance system. They are assigned based on the alternatives' estimated performance, where the estimates are obtained through replicating the model. Therefore, it is important that an accurate estimate is obtained in order to minimize the occurrence of a wrongful decision about the direction that the search should continue. Furthermore, it is important to obtain this estimate without performing a large number of replications. This research develops an indifference-zone ranking and selection procedure under common random numbers to overcome this problem. By using an indifference-zone procedure, a statistical guarantee can be made about the direction the search should progress as well as a statistical guarantee about the results from the search. Using CRN significantly reduces the required number of replications.; The proposed simulation optimization algorithm was applied to four case studies. Three of those case studies have frequently been used in industrial engineering literature for comparing the performance of proposed optimization algorithms and statistical comparison procedures. Results from these case studies indicate that the proposed optimization algorithm outperforms tabu search, simplex algorithm, and Scenario Seeker. Furthermore, it performs at least as well as evolutionary strategies. Results also indicate that the proposed indifference-zone procedure requires far less replications than other procedures to perform the correct selection. In the fourth case study, the proposed optimization algorithm was applied to a large-scale real world simulation model. Results indicates that the proposed algorithm found good solutions while searching only a small portion of the search space.