Optimal predictive designs for experiments that involve computer simulators

Optimal predictive designs for experiments that involve computer simulators

Bibliographic Details
Main Author: Leatherman, Erin Rae
Language:English
Published: The Ohio State University / OhioLINK 2013
Subjects:
Statistics
experimental design
computer experiments
prediction
calibration
computer simulator
Online Access:http://rave.ohiolink.edu/etdc/view?acc_num=osu1376393067
id ndltd-OhioLink-oai-etd.ohiolink.edu-osu1376393067
record_format oai_dc
spelling ndltd-OhioLink-oai-etd.ohiolink.edu-osu13763930672021-08-03T06:19:10Z Optimal predictive designs for experiments that involve computer simulators Leatherman, Erin Rae Statistics experimental design computer experiments prediction calibration computer simulator A deterministic computer simulator is the numerical coding of a mathematical model that describes the relationship between variables in a physical system. Computer experiments use simulators as experimental tools to determine "responses" or "outputs" at a set of user-specified input design points. As mathematical descriptions have become more sophisticated, the use of deterministic simulators as experimental vehicles has become more widespread in many scientific applications. In fact, computer experiments can sometimes be used instead of traditional physical experiments when the number of experimental factors are too numerous to study via a physical experiment, or when a physical experiment is financially prohibitive or unethical to perform. Unfortunately the mathematical models used for computer experiments can be expensive to evaluate and may require days or months for the simulator to make a single run. When this is the case, the simulator output is often modeled as a Gaussian Stochastic Process (GaSP) using training data collected from the simulator at a user-specified set of input design points. When it is of interest to predict simulator output over the entire input space, classical design criteria for computer experiments select designs that are space-filling. That is, the design points are selected to be well-spread within the input space.This dissertation investigates an alternative, process-based design criterion for prediction which minimizes the Bayesian Integrated Mean Square Prediction Error (BIMSPE). The BIMSPE is calculated by averaging the Integrated Mean Square Prediction Error across a prior distribution for the model parameters. This dissertation uses the minimum BIMSPE criterion to find simulator designs that allow for good prediction of the simulator output. The predictive ability of the BIMSPE-optimal designs are compared to IMSPE-optimal designs and to space-filling designs using a simulation study.When computer simulators are able to describe the true mean of an associated physical experiment, physical observations can be used to calibrate the simulator model. Calibration is performed so that the simulator output is as "close as possible" to the mean physical response. This dissertation also uses the minimum BIMSPE criterion to find combined physical and simulator designs that allow for good prediction of the true mean physical response using the calibrated simulator. Using a simulation study, the predictive ability of the combined BIMSPE-optimal designs are compared to combined IMSPE-optimal designs and classic physical and simulator designs that are used for prediction. This dissertation shows that in both the simulator-only setting and the calibration setting, the classic designs perform much worse than the IMSPE- and BIMSPE-optimal designs. 2013-12-19 English text The Ohio State University / OhioLINK http://rave.ohiolink.edu/etdc/view?acc_num=osu1376393067 http://rave.ohiolink.edu/etdc/view?acc_num=osu1376393067 unrestricted This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws.
collection NDLTD
language English
sources NDLTD
topic Statistics
experimental design
computer experiments
prediction
calibration
computer simulator
spellingShingle Statistics
experimental design
computer experiments
prediction
calibration
computer simulator
Leatherman, Erin Rae
Optimal predictive designs for experiments that involve computer simulators
author Leatherman, Erin Rae
author_facet Leatherman, Erin Rae
author_sort Leatherman, Erin Rae
title Optimal predictive designs for experiments that involve computer simulators
title_short Optimal predictive designs for experiments that involve computer simulators
title_full Optimal predictive designs for experiments that involve computer simulators
title_fullStr Optimal predictive designs for experiments that involve computer simulators
title_full_unstemmed Optimal predictive designs for experiments that involve computer simulators
title_sort optimal predictive designs for experiments that involve computer simulators
publisher The Ohio State University / OhioLINK
publishDate 2013
url http://rave.ohiolink.edu/etdc/view?acc_num=osu1376393067
work_keys_str_mv AT leathermanerinrae optimalpredictivedesignsforexperimentsthatinvolvecomputersimulators
_version_ 1719434662892273664

Similar Items