A best-first search approach for determining starting regions in simulation optimization

Many simulation optimization approaches assume that a "good" starting point is identified, that the design grid (i.e., spacing of runs for searching) is known, and the only one basic search method need be employed. Often, how ever, one or more of these items is unknown or is inappropriate. These assu mptions can lead to an unnecessary expenditure of simulation runs, failure to find the simulation optimum, and/or a false declaration of the optimal c onditions. This article proposes that an approach based on best-first searc h be used to determine the optimization starting region, starting point, an d design grid. The best-first search works in concert with a divide-and-con quer strategy for partitioning the search space and a safety net that acts as a conservative check to prevent permanent pruning of desirable regions. The methodology is demonstrated and shown to be successful on three example problems.