Popis: |
Discrete event simulation is currently the primary analysis tool for manufacturing systems. However, it must be link with an optimization technique to be effectively used for designing new systems. The challenge in simulation optimization is to find an appropriate technique for selecting the various inputs that represent the optimal configuration. This research compared four promising simulation optimization algorithms by testing each algorithm on twenty realistic problems. The four selected simulation optimization algorithms are: pattern search, simplex, simulated annealing and genetic algorithms. The criteria for evaluating the tested algorithms include: solution value, computation time, and type of problem solved. As a result, on average pattern search and genetic were closer to the best solution found than the other tested algorithms. For the small test problems, the pattern search is the best algorithm. For the large test problems, the genetic algorithm outperformed the pattern search significantly. For medium problems, the average solution of the genetic algorithm is about five percent better than the average pattern search solution. However, the average number of replications of the genetic algorithm is about three time higher than the pattern search. Therefore, one has to evaluate the trade off between pattern search with a smaller number of replications and the genetic algorithm with better results. |