SIMULATION METAMODELING AND OPTIMIZATION WITH AN ADDITIVE GLOBAL AND LOCAL GAUSSIAN PROCESS MODEL FOR STOCHASTIC SYSTEMS

Abstract

Many computer models of large complex systems are time consuming to experiment on. Even when surrogate models are developed to approximate the computer models, estimating an appropriate surrogate model can still be computationally challenging. In this thesis, we propose an Additive Global and Local Gaussian Process (AGLGP) model that is a flexible surrogate for stochastic computer models. The proposed additive structure of the model reduces the computational complexity in model fitting, and allows for more efficient predictions with large data sets. We show that this model form is effective in modeling various complicated stochastic computer models. With its global and local structure, we integrate the AGLGP model into a combined global and local optimization (CGLO) algorithm and show the performance and properties of the CGLO algorithm. We also extend the CGLO into a parallel framework with parallel local search to further improve the computational efficiency.