Publication: Layer-recurrent network in identifying a nonlinear system
Date
2008
Authors
Nordin F.H.
Nagi F.H.
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Layer-Recurrent Network (LRN) is a dynamic neural network and is seen as a promising black box model in identifying a nonlinear system injected with nonlinear input signal. In this paper, LRN will be used to identify a nonlinear, state space 3-axis satellite model. Open loop identification is applied and methodology on nonlinear system identification is presented where the best pair of input and output data is first measured. Using the simulated data, six LRN models are used to identify the satellite dynamics. It is shown that only 200 epochs are needed to train a network to converge to a reasonable mean squared value (mse). LRN output is then compared with the state space model where it shows that LRN model is capable to produce similar results as the state space satellite model without knowing the system's state and prior knowledge of the system.
Description
Keywords
Layer-Recurrent Network (LRN) , Nonlinear input , Nonlinear system identification , Satellite attitude , Mathematical models , Metropolitan area networks , Navigation , Neural networks , Nonlinear systems , Satellites , Black box models , Dynamic Neural networks , Input and outputs , Layer-Recurrent Network (LRN) , Mean-squared values , Nonlinear input , Nonlinear system identification , Open loop identifications , Prior knowledges , Recurrent networks , Satellite attitude , Satellite dynamics , Satellite models , Simulated datums , State Space models , State spaces , Identification (control systems)