This research investigates the integration of evolutionary techniques for symbolic regression. In particular the genetic programming paradigm is used together with other evolutionary computational techniques to develop novel approaches to the improvement of areas of simple preliminary design software using empirical data sets. It is shown that within this problem domain, conventional genetic programming suffers from several limitations, which are overcome by the introduction of an improved genetic programming strategy based on node complexity values, and utilising a steady state algorithm with subpopulations. A further extension to the new technique is introduced which incorporates a genetic algorithm to aid the search within continuous problem spaces, increasing the robustness of the new method. The work presented here represents an advance in the Geld of genetic programming for symbolic regression with significant improvements over the conventional genetic programming approach. Such improvement is illustrated by extensive experimentation utilising both simple test functions and real-world design examples.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:481293 |
| Date | January 1999 |
| Creators | Watson, Andrew Harry |
| Publisher | University of Plymouth |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/10026.1/1669 |
Page generated in 0.0021 seconds