To consider the problem of efficiently allocating simulation replications in order to maximize the probability of selecting the best design. Optimal allocation conditions are proposed for a two design case, and it is found that in the presence of a positive correlation, unless the variance of one design is considerably larger than the other, the number of simulation r replications are identical. Extending this algorithm to a general number of competing designs, an approximation is obtained for the asymptotically optimal allocation. The algorithm is tested on several numerical examples. (16 refs.)
展开▼
