Optimization of ann applied to non-linear system identification

Authors:
Oswaldo Ludwig Júnior;Pablo Corral Gonzalez;A. C. de C. Lima
Affiliations:
Faculdade de Tecnologia e Ciência - FTC, Department of Mechatronic Engineering, Salvador-BA-Brazil;Universidad Miguel Hernández - UMH, Signal Theory and Communications, Elche-Alicante-Spain;Universidade Federal da Bahia - UFBA, Department of Electric Engineering, Salvador-BA-Brazil
Venue:
MIC'06 Proceedings of the 25th IASTED international conference on Modeling, indentification, and control
Year:
2006

Citing 2
Cited 2

Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks

Neural Networks
Practical genetic algorithms

Practical genetic algorithms

Novel maximum-margin training algorithms for supervised neural networks

IEEE Transactions on Neural Networks
Eigenvalue decay: A new method for neural network regularization

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this work is we propose a hybrid technique for artificial neural network optimization. This technique is a combination of Real-Coded Genetic (GA) and Back Propagation (BP) algorithms. The main idea is to apply the Back Propagation over the five more adapted individuals of each generation produced by the Real Coded GA, in order to allow the evolution of these individuals before the crossover operation. The combined algorithms are used in non-linear system identification. Some solutions are presented to solve common problems arisen from application of Genetic Algorithm. To illustrate the use of the technique we present an example of non-linear model identification.