Adaptations et applications de modèles mixtes de réseaux de neurones à un processus industriel

Abstract : This study is interested in analyzing the contribution of artificial neural networks in order to improve the control of complex industrial processes that are mainly characterized by their temporal behavior. The main motivations of the time series analysis are data reduction, indexation based on similarity, localization of sequences, knowledge extraction and prediction. The analyzed industrial process is an electric arc furnace for the liquid steel production in Luxembourg. The proposed approach is a concept of predictive control based on unsupervised learning techniques with the aim of knowledge extraction. Our signal coding method is based on primitive patterns that compose the signals. These patterns, building the coding alphabet, are extracted using an unsupervised method, the self organizing maps of Kohonen (SOM). An alphabet validation approach is proposed. One of the important subjects of this research is the similarity of time series. The proposed method is unsupervised and able to handle sequences of arbitrary size using the Dynamic Time Warping method (DTW).
Document type :
Theses
Complete list of metadatas

https://hal.univ-lorraine.fr/tel-01748184
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 11:27:43 AM
Last modification on : Tuesday, April 24, 2018 - 1:53:22 PM
Long-term archiving on: Friday, September 14, 2018 - 8:46:19 AM

File

SCD_T_2006_0189_SCHUTZ.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01748184, version 1

Collections

Citation

Georges Schutz. Adaptations et applications de modèles mixtes de réseaux de neurones à un processus industriel. Autre [cs.OH]. Université Henri Poincaré - Nancy 1, 2006. Français. ⟨NNT : 2006NAN10189⟩. ⟨tel-01748184⟩

Share

Metrics

Record views

26

Files downloads

28