Stochastic Kriging: Modeling and Controlling Uncertainty in Simulation

Northwestern University, Evanston IL

Investigators

Abstract

PI: PI_NAME Staum, Jeremy INSTITUTION: Northwestern University TITLE: Stochastic Kriging: Modeling and Controlling Uncertainty in Simulation "Stochastic Kriging: Modeling and Controlling Uncertainty in Simulation" The research aims to improve algorithms that rely on simulation of stochastic systems to support decision-making by increasing the algorithms? computational efficiency and enabling more reliable statistical inference about the quality of proposed decisions. Two kinds of algorithms are considered. One kind is for the exploration and optimization of the performance of a stochastic system as a function of its design. The other kind is for assessing risks due to uncertainty about system parameters, e.g. the average time required to service a customer, or about financial scenarios, involving e.g. asset prices or loan defaults. The primary research objective is to improve the algorithms by devising efficient plans for running simulations that can support accurate predictions of system performance given any of a wide range of designs. Improvements to algorithms for the exploration and optimization of the performance of a stochastic system as a function of its design could make it possible to provide decision-makers instantly with accurate estimates of the performance of system designs they wish to explore. This would enable simulation to better support decisions about investment in manufacturing and service systems. Better algorithms for risk assessment would encourage better decision-making and risk management practices. At present, risks due to uncertainty about system parameters or about financial scenarios are often treated with methods less accurate or less illuminating than simulation, because simulation is thought to be too unwieldy or slow to handle those risk assessment problems.