An Improved Budget Allocation Procedure for Selecting the Best-Simulated Design in the Presence of Large Stochastic Noise

Abstract

Motivated by the increasing practical needs for simulation optimization of modern industrial systems, this paper proposes an efficient ranking and selection (R&S) procedure for selecting the best-simulated design from a finite set of alternatives in the presence of large stochastic noise. To obtain the correct selection under a limited simulation budget, the proposed procedure sequentially allocates the budget to minimize the evaluated uncertainty values of the selection through a two-step process based on the existing uncertainty evaluation (UE) procedure. This two-step process reduces the inefficiency of the underlying UE procedure while keeping its high robustness to noise, thereby achieving improved the efficiency for the proposed procedure in a noisy environment. This improved efficiency is demonstrated in comparative experiments with other R&S procedures on several benchmark problems. In particular, the experimental results of three practical optimization problems emphasize the necessity of the proposed procedure.