Skip to Main content Skip to Navigation
Theses

Méthodes d'optimisation multimodales associées à la modélisation numérique en électromagnétisme

Abstract : In recent years, expansion of computer sciences and artificial intelligence led to a unprecedented development of optimization methods which are today capable of taking into account a great number of design variables. In particular, evolutionary algorithms follow an exponential growth since the beginning of the sixties and are progressively considered as the most robust optimization techniques. On the one hand, they converge to an optimum of a given objective function without using any partial derivatives of this function in relation to design variables. On the other hand, they are not trapped by local optima and often succeed in determining the global optimum. However, translating a real optimization problem in a mathematical form, by taking into account several characteristics (efficiency, sensitivity, cost price,...), is not easy. Consequently, engineers appreciate when they study one aspect to get various "good" possibilities rather than a single solution. For that purpose, niching genetic algorithms offer interesting perspectives by locating multiple optimal solutions either local as global. Our work is centered on the characterization of these new optimization techniques. Each niching method is investigated on mathematical test functions and on electromagnetic benchmarks. Comparisons are also carried out on an original process of automatic electrode shape design. We describe a new approach for 2D and axisymetric devices where the electrode shape is identified from an equipotential obtained from a system of fictious point charges.
Complete list of metadata

Cited literature [50 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00139202
Contributor : Publications Ampère <>
Submitted on : Thursday, March 29, 2007 - 5:58:18 PM
Last modification on : Wednesday, March 31, 2021 - 3:10:05 AM
Long-term archiving on: : Wednesday, April 7, 2010 - 1:46:59 AM

Identifiers

  • HAL Id : tel-00139202, version 1

Citation

Bruno Sareni. Méthodes d'optimisation multimodales associées à la modélisation numérique en électromagnétisme. Autre. Ecole Centrale de Lyon, 1999. Français. ⟨tel-00139202⟩

Share

Metrics

Record views

783

Files downloads

3447