Abstract
Simulation optimization is one kind of optimization methods aimed to find the best solution in a simulated stochastic system. Especially in PC-era, simulation optimization has been attracting a lot of attention, and adopted in many practical problems. However, classical simulation optimization methods focused on expectation-based problems; seldom researches considered quantile-based problems. In this thesis, a gradient-based framework for quantile-based simulated optimization (GBQS) has been proposed. GBQS is based on the framework of STRONG-S, and modified it to fit the quantile-based case. GBQS is designed to solve not only lower dimensional problems but also higher ones. For efficiency and controlling solution qualification purpose, GBQS uses a good deal of statistical techniques such as design of experiments, quantile regression, factor screening, and hypothesis testing. GBQS is verified as a highly adaptable method because it has good performance on many situations by testing for several numerical experimental problems and a practical problem.