OPTIMAL COMPUTING BUDGET ALLOCATION FOR STOCHASTIC SIMULATION OPTIMIZATION

Abstract

Optimal computing budget allocation (OCBA) considers the allocation of limited simulation budget based on stochastic simulation output in order to optimize the probability of correct selection. It is not only a promising ranking and selection procedure for small-scale simulation optimization problems, but also an effective simulation allocation framework when incorporated with search algorithms for large-scale problems. In this thesis, we extend the framework of OCBA in various aspects to efficiently solve problems with different objectives, scales, and distribution assumptions. From the perspective of problem setting, we extend the objective function in various ways. We also consider non-Gaussian distributions for simulation output. From the practitioner’s perspective, we derive easy-to-implement simulation allocation rules among all configurations. For multi-objective problems of large-scale, we demonstrate the integration of OCBA with a new multi-objective particle swarm optimization algorithm to optimize the stochastic search process. We also propose a general multi-objective simulation optimization framework to further improve the sampling efficiency of medium-scale problems when simulation models with multiple fidelity levels exist.