A tutorial on kriging-based stochastic simulation optimization

TL;DR

This tutorial explains how kriging models can be used for efficient simulation optimization, especially when simulations are costly and noisy, by comparing methods and exploring extensions for complex problems.

Contribution

It provides a comprehensive overview of kriging-based algorithms, their advantages over traditional methods, and discusses extensions for multi-objective and constrained optimization.

Findings

Kriging-based algorithms reduce the number of simulations needed.

They outperform polynomial-based methods in noisy, expensive simulations.

Extensions enable handling complex, real-world optimization problems.

Abstract

This tutorial focuses on kriging-based simulation optimization, emphasizing the importance of data efficiency in optimization problems involving expensive simulation models. It discusses how kriging models contribute to developing algorithms that minimize the number of required simulations, particularly in the presence of noisy evaluations. The tutorial compares the performance of kriging-based algorithms against traditional polynomial-based optimization methods using an illustrative example. Additionally, it discusses key extensions of kriging-based algorithms, including multi-objective and constrained optimization, providing insights into their application in complex, real-world settings.