Process control, optimization, and modeling of chemical systems using genetic algorithms and neural networks /
Genetic Algorithms (GAs) are emerging as powerful alternatives to traditional optimization methods which are too restrictive and CPU intensive. The first part of this thesis is aimed at studying the use of GAs to solve optimization and search problems in chemical engineering. The simple genetic alg...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1995.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=742534851&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | Genetic Algorithms (GAs) are emerging as powerful alternatives to traditional optimization methods which are too restrictive and CPU intensive. The first part of this thesis is aimed at studying the use of GAs to solve optimization and search problems in chemical engineering. The simple genetic algorithm is modified by use of a clustering algorithm to identify and dissolve clusters so that premature convergence is prevented while searching for global optimum. A modification of the genetic search procedure to incorporate qualitative objective functions is also proposed. This helps the optimization of controllers when the figure of merit is hard to express by mathematical expressions. The ability of a GA to conveniently solve a mixed integer nonlinear programming (MINLP) problem is exploited to solve controller synthesis and design problems. A hierarchical decomposition of the controller design problem is proposed which helps us to formulate the controller design problem as an MINLP problem which can be conveniently solved by a GA. In the second part of this thesis, the use of recurrent neural networks in process control of nonlinear systems is investigated. An integrated methodology is presented, for the modeling and controller design of nonlinear dynamical systems. An RNN is trained using plant input-output data and is used as an observer in the design of an exact-linearizing controller. The controller rises the states of the RNN and measurement of plant states is not required. The methodology was tested on both SISO and MIMO systems and shown to perform better than a linear, optimally tuned controller In the third part of this thesis, the effects of system nonlinearity on feedback controller design are studied. Nonlinearities can be usefully quantified by the 2-norm for various ranges of process inputs and this knowledge can be used to explain why and when linear controllers are adequate. We demonstrate that a nonlinear system may or may not necessitate the use of a nonlinear model for control purposes. Given the significant effort required to design and maintain nonlinear controllers a quantitative analysis of the need for such an approach would be a worthwhile task to complete before nonlinear control is attempted. |
|---|---|
| Item Description: | Vita. "Major Subject: Chemical Engineering". |
| Physical Description: | xxv, 214 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilms Inc. |
| Bibliography: | Includes bibliographical references. |