Skip to Main content Skip to Navigation
Habilitation à diriger des recherches

Approches neuromimétiques pour l'identification et la commande

Abstract : New adaptive and neural strategies for the identification and the control of complex, non-linear and non-stationary systems are developed in this research work. Artificial neural networks are known as neural models or connectionism. They are endowed with universal approximation and are able to learn from and to adapt to their environment with weak assumptions. These attributes make them interesting from an engineering perspective. They significantly improve not only identification and control schemes themselves but also the way in which they can be used for considering a system and its interaction with its environment. The present study aims at developing new neural schemes by introducing a priori knowledge in an explicit way to be closer to the considered system and to make identification and control tasks more efficient. Several developments are presented implicating formal neurons, neural architectures and neural strategies. A formal neuron has been optimized. Different modular neural approaches based on several neural networks have been proposed. Neural schemes resulting from a theoretical analysis of systems have been introduced. This formalism lies on the mathematical expression of internal system signals and uses synthesized signals representing its evolution. Combining artificial neural networks with techniques such as fuzzy logic, statistical models or other parametric models have also been investigated. The proposed neural techniques have been experimentally assessed. We have shown that neural network approaches resulting from a thoughtful design strategy act for advanced and efficient controllers.
Document type :
Habilitation à diriger des recherches
Complete list of metadatas

Cited literature [237 references]  Display  Hide  Download
Contributor : Patrice Wira <>
Submitted on : Wednesday, July 6, 2011 - 7:32:48 PM
Last modification on : Friday, October 23, 2020 - 4:41:52 PM
Long-term archiving on: : Friday, October 7, 2011 - 2:20:25 AM


  • HAL Id : tel-00605218, version 1



Patrice Wira. Approches neuromimétiques pour l'identification et la commande. Sciences de l'ingénieur [physics]. Université de Haute Alsace - Mulhouse, 2009. ⟨tel-00605218⟩



Record views


Files downloads