Files in this item



application/pdf9215817.pdf (5MB)Restricted to U of Illinois
(no description provided)PDF


Title:Equation-based chemical process optimization on advanced architecture computers
Author(s):Goheen, Christopher Hembrough
Doctoral Committee Chair(s):Stadtherr, Mark A.
Department / Program:Chemical and Biomolecular Engineering
Discipline:Chemical and Biomolecular Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Engineering, Chemical
Abstract:The equation-based approach shows great promise as an optimization tool, but its use has been hindered by implementation difficulties. These include problems in providing good automatic initial guesses and excessive computational time on large problems. A prototype chemical process optimization package, SEQUEL-III, is used to consider these issues.
The successive quadratic programming method is applied for the solution of large-scale nonlinear programming problems arising from the equation-based optimization of chemical processes. Since the algorithm updates a full approximate Hessian matrix, a large central memory and vectorization are required for this implementation. The reliability of the program is enhanced with an evolutionary initialization scheme and automatic generation of physically reasonable bounds on the variables. SEQUEL-III has solved a variety of test problems, ranging in size to nearly 1500 variables. For these problems, the optimization takes 2 to 20 times as long as the solution of the related simulation or design problem.
The structure of the occurrence matrix of the Hessian of the Lagrangian is used to consider the sparsity implications involved in solving large-scale problems. This requires that the sparsity be preserved by an augmented Lagrangian or sparse updating technique. In either case, a two-pass approach can be applied for solving the large, sparse linear systems arising at each iteration of the optimization solver. The occurrence matrix is reordered once, then the permuted matrix is solved numerically. Experiments on linear systems generated by an augmented Lagrangian optimization solver show that linear solution techniques which exploit vectorization, positive-definiteness, symmetry, sparsity, or a combination of these features achieve the greatest efficiency, solving some of the larger linear systems on a Cray-2 computer up to 30 times faster than a full solver which does not exploit these features.
Issue Date:1992
Rights Information:Copyright 1992 Goheen, Christopher Hembrough
Date Available in IDEALS:2011-05-07
Identifier in Online Catalog:AAI9215817
OCLC Identifier:(UMI)AAI9215817

This item appears in the following Collection(s)

Item Statistics