Skip to Main content Skip to Navigation
Theses

Apprentissage autonome de réseaux de neurones pour le pilotage en temps réel des systèmes de production basé sur l'optimisation via simulation

Abstract : The real-time control of production systems requires complex decisions on resource allocation and the choice of tasks to perform. Given the importance of the relevance of decisions for the performance of a workshop on the subject of steering research whose goal is to help decision makers. In particular, it is unclear assess the impact on the performance of a real-time decision for the good performance result of a sequence of decisions, not just one. Thus, it is difficult to establish what is the best decision at any given moment. Several authors have used simulation to learn best practices using machine learning approaches but have encountered difficulty in obtaining samples or observations on real-time decisions, where the consideration state changes is essential to choose production strategies. We managed to address this problem by proposing a learning approach using neural networks, which requires no examples, observations or prior expert knowledge. This type of learning is achieved by optimization via simulation parameters of the neural network compared to a target system performance. It seeks to extract independently of knowledge on the best way to decide on a simulation model. We show faisablité and the contribution of our approach on two examples based on the literature.
Complete list of metadatas

Cited literature [83 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00725259
Contributor : Camille Meyer <>
Submitted on : Monday, August 27, 2012 - 11:06:13 AM
Last modification on : Friday, October 23, 2020 - 4:49:24 PM
Long-term archiving on: : Wednesday, November 28, 2012 - 3:30:58 AM

Identifiers

  • HAL Id : tel-00725259, version 1

Citation

Wiem Mouelhi-Chibani. Apprentissage autonome de réseaux de neurones pour le pilotage en temps réel des systèmes de production basé sur l'optimisation via simulation. Réseau de neurones [cs.NE]. Université Blaise Pascal - Clermont-Ferrand II, 2009. Français. ⟨NNT : 2009CLF21959⟩. ⟨tel-00725259⟩

Share

Metrics

Record views

778

Files downloads

5263