|M.Sc Student||Grosman Benyamin|
|Subject||Nonlinear System Modeling Using Genetic Programming|
|Department||Department of Chemical Engineering||Supervisor||Professor Daniel Lewin|
Genetic Programming is one of the computer algorithms in the family of evolutionary-computational methods. The basic idea behind all such algorithms is Darwin’s law of nature, or “the survival of the fittest”. These algorithms have been used to solve several problems in a variety of areas; in particular they excel in the solution of complex optimization problems. The Genetic Programming under discussion in this work relies on tree-like building blocks, and thus supports process modeling with varying structure.
In the present work, a modified GP code is developed to provide modeling support for a number of process system engineering (PSE) applications. It is shown that the basic algorithm facilitates the generation of steady state nonlinear empirical models for process analysis and optimization, as well as discrete dynamical ones for control applications. This work can be seen as an evolution of several works in the field and its main goal is to improve the approach efficiency and accuracy. The improved GP code includes some novel ideas about fitness calculations, creation of new generations, constants placement and parametric optimization. The advantages of these changes are tested against the more commonly-used approaches. The work is divided in two main parts. In the first, the creation of steady state models is examined, while the second part examines the development of discrete dynamic models, a procedure that demands special treatment. The derived discrete dynamic models are then implemented in a model predictive control scheme. This implementation can stand-alone as a unique model predictive control approach or as a part of the general work done on the GP code by emphasizing its abilities.