Título: Aplicação de Redes Neurais RBF Treinadas com Algoritmo ROLS e Análise Wavelet na Predição de Tráfego em Redes Ethernet
Autores: Vieira, Flávio Henrique Teles; Lemos, Rodrigo Pinto
Resumo: The use of a wavelet transform technique is investigated for Ethernet traffic prediction by RBF (Radial Basis Function) neural networks based on the combination of the signal approximation and detail predictions in different resolution levels. Then, useful information is extracted resulting in a more appropriate modelling of non-stationary time series by the RBF neural network. The ROLS (Regularised Orthogonal Least Squares) algorithm is applied for the neural networks training to produce smaller nets regarding the number of centers. This contributes to avoid problems of bad conditioning as well as to give an extrapolation capacity improvement verified by the mean-square error reduction of the prediction.
Código DOI: 10.21528/CBRN2003-034
Artigo em PDF: 6CBRN_034.PDF
Arquivo BibTex: 6CBRN_034.bib