Title :
Extrapolative models of dynamics systems: neural networks vs. Kalman filter
Author :
Mu´awin, A. ; Chowdhury, Fahmida N.
Author_Institution :
Dept. of Electr. Eng., Michigan Technol. Univ., Houghton, MI, USA
Abstract :
The authors present a case study for comparing two modelling techniques for nonlinear systems: neural networks and ARMA (autoregressive moving average). The ARMA model is generated on-line by a Kalman filter, and the coefficients are allowed to be slowly time-varying. The neural network has one hidden layer with sigmoid neurons, and a time-delay structure. Simulations results are presented
Keywords :
Kalman filters; autoregressive moving average processes; delays; extrapolation; industrial plants; modelling; neural nets; nonlinear systems; simulation; ARMA; Kalman filter; autoregressive moving average; dynamics systems; extrapolative models; hidden layer; modelling techniques; neural networks; nonlinear systems; sigmoid neurons; simulation; slowly time-varying coefficients; time-delay structure; Autoregressive processes; Delay; Extrapolation; Neural networks; Neurons; Nonlinear equations; Nonlinear systems; Predictive models; Testing; Training data;
Conference_Titel :
System Theory, 1997., Proceedings of the Twenty-Ninth Southeastern Symposium on
Conference_Location :
Cookeville, TN
Print_ISBN :
0-8186-7873-9
DOI :
10.1109/SSST.1997.581648
