Accession Number:

ADA465291

Title:

Surrogate Structures for Computationally Expensive Optimization Problems With CPU-Time Correlated Functions

Descriptive Note:

Master's thesis

Corporate Author:

AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING AND MANAGEMENT

Personal Author(s):

Report Date:

2007-06-01

Pagination or Media Count:

74.0

Abstract:

This research focuses on numerically solving a class of computationally expensive optimization problems that possesses a unique characteristic as the optimal solution is approached, the computational time required to compute an objective function value decreases. This is motivated by an application in which each objective function evaluation requires both a numerical uid dynamics simulation and an image registration and comparison process. The goal is to nd the parameters of a predetermined image by comparing the ow dynamics from the numerical simulation and the predetermined image through the image comparison process. The generalized pattern search and mesh adaptive direct search methods were applied in a way that employs surrogate functions in the search step to reduce the number of costly function evaluations. The surrogate functions are formed, based on either previous function values or their computational times, or both. The solution to the surrogate optimization problem can be solved easily and provides an improved solution quickly. A time cut-o parameter was also added to the objective function to allow its termination during the comparison process if the computational time exceeded a specied threshold. The approach was tested on two problems using the NOMADm and DACE MATLABr software packages, and results are presented.

Subject Categories:

  • Numerical Mathematics
  • Computer Programming and Software
  • Computer Systems

Distribution Statement:

APPROVED FOR PUBLIC RELEASE