Title :
From batch to recursive outlier-robust identification of non-linear dynamic systems with neural networks
Author :
Thomas, P. ; Bloch, G.
Author_Institution :
Centre de Recherche en Autom., Vandoeuvre, France
Abstract :
The problem of identification for nonlinear SISO systems in the presence of outliers in data is considered. Neural networks are used for their capabilities to solve nonlinear problems by learning. Three prediction error learning rules based on outlier-robust criteria are drawn up, for batch and recursive identification. The robust recursive algorithms are compared with the standard Levenberg-Marquardt update rule through a simulation example of fault detection
Keywords :
learning (artificial intelligence); neural nets; nonlinear dynamical systems; prediction theory; recursive estimation; batch outlier-robust identification; fault detection; neural networks; nonlinear SISO systems; nonlinear dynamic systems; prediction error learning rules; recursive outlier-robust identification; standard Levenberg-Marquardt update rule; Artificial neural networks; Backpropagation algorithms; Fault detection; Fault diagnosis; Feedforward neural networks; Multi-layer neural network; Neural networks; Neurons; Robustness; System identification;
Conference_Titel :
Neural Networks, 1996., IEEE International Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-3210-5
DOI :
10.1109/ICNN.1996.548887
