Paper:
Language: English:

Evolutionary Nonlinear Multimodel Partitioning Filters

G. N. Befigiannis^*,**, E. N. Demiris^* and S. D. Likothanassis^,,

Department of Computer Engineering and Informatics University of Patras, Rio, 26500, Patras, Greece
Computer Technology Institute (C.T.I.) Kolokotroni 3, 26221, Patras, Greece
University of Patras Artificial Intelligence Research Center (U.P.A.I.R.C.) University of Patras, Rio, 26500 Patras, Greece

Received: October 20, 2000

Accepted: December 10, 2000

Keywords: Nonliner filter, Multimodel Partitioning Theory, Genetic Algorithms

Journal ref: Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol.5, No.1 pp. 8-14, 2001

Abstract

The problem of designing adaptive filters for nonlinear systems is faced in this work. The proposed evolution program combines the effectiveness of multimodel adaptive filters and the robustness of genetic algorithms (GAs). Specifically, a bank of different extended Kalman filters is implemented. Then, the a posteriori probability that a specific model of the bank of conditional models is the true one can be used as a GA fitness function. The superiority of the algorithm is that it evolves concurrently the models\' population with initial conditions. Thus, this procedure alleviates extended Kalman filter sensitivity in initial conditions, by estimating the best values. In addition to this, adaptive implementation is proposed that relieves the disadvantage of time-consuming GA implementation. Finally, a variety of defined crossover and mutation operators is investigated in order to accelerate the algorithm\'s convergence.

[Notice]
* "Preview" is the first 2 pages of the article. You don't need the registration.
* To read the PDF file you will then need to download and install the Adobe Reader.
Adobe Reader is free and available for download here: