Contribution à l'identification de systèmes non linéaires par réseaux de neurones

Abstract : This thesis deals with the idenlification of dynamical non-tinear MISO Sytems with multilayer feedfoward neural networks. Firstly, a short presentation of the non-linear identification methods is proposed and the neural network used is more precesialy defined. Some methods are presented to adapt his architecture to a particular case. These methods give the regressors and the number of neurons in the hidden layer. The relationships between neural identification and the most dassical non-linear models are then hown. The validation criteria of non-tineal' models usable for the neural identification are presented. Three difficulties encountered in neural identification are investigated in the sequel. The first one is due to the iniatilisation of the parameters of the network. A bad choice of these initial parameters can lead to local minima very far from of the global minimum, to saturation of the hidden neurons, or to slow convergence. Two new algorilhm are proposed to solve this problem and compared wilh others on three different examples. The slow convergence can be the result of the learning algorithm used. One algorithm is proposed to deal witgh this second difficulty. This algorithm is compared with the more dassical RPE algorithm. This study is ends with the third studied problem which is posed by the presence or outliers in the identification data set. lndeed, outliers can produce biases on estimaled parameters.Three robust criteria are then proposed and are compared with the classical quadratic criterion on a simulation example and on a real industrial data set.
Document type :
Theses
Complete list of metadatas

https://hal.univ-lorraine.fr/tel-01747429
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 10:59:45 AM
Last modification on : Saturday, April 14, 2018 - 1:28:28 AM
Long-term archiving on : Friday, September 14, 2018 - 12:26:55 AM

File

SCD_T_1997_0030_THOMAS.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01747429, version 1

Collections

Citation

Philippe Thomas. Contribution à l'identification de systèmes non linéaires par réseaux de neurones. Autre. Université Henri Poincaré - Nancy 1, 1997. Français. ⟨NNT : 1997NAN10030⟩. ⟨tel-01747429⟩

Share

Metrics

Record views

60

Files downloads

132