Performance of feasible Markov chain-based predictors for nonlinear systems

Author

Gao, H. ; Karny, M. ; Slama, M.

Author_Institution

Dept. of Adaptive Syst., Inst. of Inf. Theor. & Autom., Prague, Czech Republic

fYear

1997

fDate

1-7 July 1997

Firstpage

2282

Lastpage

2287

Abstract

A non-traditional adaptive predictor is successfully compared with a neural network. It combines several simple Markov chain-based predictors gained from Bayesian estimation with forgetting. It can describe non-linear. stochastic digitized dynamic systems with finite memory and slowly varying parameters. Its complexity is linear in the number of used models m and the number of input levels m_u. and quadratic in the number of output levels m_y. This is in sharp contrast with the corresponding full Markov predictor whose complexity is (m_ym_u)^m+1.

Keywords

Bayes methods; Markov processes; adaptive control; computational complexity; neurocontrollers; nonlinear control systems; predictive control; stochastic systems; Bayesian estimation; Markov chain-based predictor; Markov predictor; finite memory; linear complexity; neural network; nonlinear system; nontraditional adaptive predictor; stochastic digitized dynamic system; varying parameter; Adaptation models; Approximation methods; Artificial neural networks; Bayes methods; Data models; Prediction algorithms; Predictive models; Adaptive; Estimation; Neural nets;

fLanguage

English

Publisher

ieee

Conference_Titel

Control Conference (ECC), 1997 European

Conference_Location

Brussels

Print_ISBN

978-3-9524269-0-6

Type

conf

Filename

7082445