Selecting a Good Enough Simulated Design with Opportunity Cost

Abstract

Statistical selection approaches are used to identify the best simulated design from a finite set of alternatives. For each alternative, stochastic simulation is used to deduce the value of performance measures. Since we used simulation to get the estimates value of performance measures, there is a potential for incorrect selection. There are two measures of selection quality; the Probability of Correct Selection and the Expected Opportunity Cost of potentially incorrect selection. In this paper, we present a combined approach for selecting a good stochastic design with high probability when the number of elements in the feasible solution set is huge. In the first stage, Ordinal Optimization approach is used to select randomly a subset that intersects with the set of the actual best k% design with high probability from the search space. The next step we used Optimal Computing Budget Allocation method to allocate the available simulation samples in a way that maximize the Probability of Correct Selection. Then we used Subset Selection approach to get a smaller subset that contains the best design from the subset that is selected before. Finally, we
used the Indifference–Zone approach to select the best design from the survivors in the previous stage. This approach is tested on numerical example, and the results show that this approach selects the best design with high probability, when the number of alternative is large. On the other hand we see from the numerical results, this approach select the desired design with the minimum Expected Opportunity Cost of a potentially incorrect selection.