Katya Vladislavleva
| Date of Ph.D. defense: | 5 September 2008 |
| Title of thesis: | Model-based Problem Solving through Symbolic Regression via Pareto Genetic Programming |
| ISBN: | 978 90 5668 217 0 |
| Promotor: | Prof.dr.ir. D. den Hertog |
| Co-promotor: | Dr. G.F. Smits |
Abstract:
This dissertation focuses on symbolic regression for data-driven modeling and system identification. The search for explicit input-response regression models which are simple, interpretable, accurate, and trustworthy is performed via Pareto genetic programming - a multi-objective variant of genetic programming evolutionary methodology designed for developing models with competing criteria of fitness.
The thesis exploits a generic framework of adaptive model-based problem solving used in industrial modeling applications. This framework consists of an iterative feed-back loop over: (Part I) data generation, analysis and adaptation, (Part II) model development, and (Part III) problem analysis and reduction.
Pareto genetic programming methodology is extended by additional generic model selection and generation strategies that (1) drive the modeling engine to creation of models of reduced non-linearity and increased generalization capabilities, and (2) improve the effectiveness of the search for robust models by goal softening and adaptive fitness evaluations.
In addition to the new strategies for model development and model selection, this dissertation presents a new approach for analysis, ranking, and compression of given multi-dimensional input-response data for the purpose of balancing the information content of undesigned data sets.
Global / English