Abstract
Simulation optimization has received a great deal of attention over the decades due to its generality and solvability in many practical problems. On the other hand, simulation optimization is well recognized as a difficult problem, especially when the problem dimensionality grows. Stochastic Trust-Region Response Surface Method (STRONG) is a newly developed method built upon the traditional Response Surface Methodology (RSM). Like the traditional RSM, STRONG employs efficient design of experiments and regression analysis; hence, it can enjoy computational advantages for higher-dimensional problems. However, STRONG is superior to the traditional RSM in that it is an automated algorithm and has provable convergence guarantee. This article exploits the structure of STRONG and proposes a new framework that combines STRONG with efficient screening designs to enable the solving of large-scale problems; e.g., hundreds of factors. It is shown that the new framework is convergent with probability one. Numerical experiments show that the new framework is capable of handling problems with hundreds of factors and its computational performance is far more satisfactory than other existing approaches. Two illustrative examples are provided to show the viability of the new framework in practical settings.