Optimal Budget Allocation Rule for Simulation Optimization Using Quadratic Regression in Partitioned Domains

Ranking and selection procedures have been successfully applied to enhance the efficiency of simulation in recent years. To further improve the efficiency, one approach is to incorporate the simulation output from across the domain into some response surfaces. In this paper, the domain of interest is divided into adjacent partitions and a quadratic regression function is assumed for the mean of the underlying function in each partition. Using the large deviation theory, an asymptotically optimal allocation rule is proposed with the objective of maximizing the probability of correctly selecting the best design point. The proposed simulation budget allocation rule is implemented in a heuristic sequential allocation algorithm and compared with some existing allocation rules. Numerical results illustrate the effectiveness of the proposed simulation budget allocation rule.