We develop a neuro-fuzzy network technique to extract TSK-type fuzzy rules from a given set of input-output data for classification problems. Fuzzy clusters are generated incrementally from the training data set, and similar clusters are merged dynamically together through input-similarity, output-similarity, and output-variance tests. The associated membership functions are defined with statistical means and deviations. Each cluster corresponds to a fuzzy IF-THEN rule, and the obtained rules can be further refined by a fuzzy neural network with a hybrid learning algorithm which combines a recursive SVD-based least squares estimator and the gradient descent method. The proposed technique has several advantages. The information about input and output data subspaces is considered simultaneously for cluster generation and merging. Membership functions match closely with and describe properly the real distribution of the training data points. Redundant clusters are combined and the sensitivity to the input order of training data is reduced. Besides, generation of the whole set of clusters from the scratch can be avoided when new training data are considered.

Orientadores: Lee Luan Ling, Flavio Henrique Teles Vieira / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-09T03:15:16Z (GMT). No. of bitstreams: 1 Sousa_LigiaMariaCarvalho_M.pdf: 2110733 bytes, checksum: 2417de66d2ca06dcb86fbce5e919906e (MD5) Previous issue date: 2007 / Resumo: O presente trabalho propõe um esquema de controle adaptativo de ?uxos baseado no modelo fuzzy TSK. Neste esquema de controle, o modelo fuzzy TSK é utilizado para prever adaptativamente o tamanho da ?la no buffer em um enlace. Com o objetivo de ajustar dinamicamente os parâmetros do modelo fuzzy TSK, propomos um algoritmo de treinamento adaptativo. Na primeira etapa do algoritmo de treinamento proposto, os parâmetros das partes premissas e das partes conseqüentes do modelo são obtidos. A segunda etapa consiste de um algoritmo de re?namento dos parâmetros do modelo baseado em gradiente descendente. A e?ciência do preditor proposto é avaliada através da comparação com outros preditores adaptativos fazendo uso de traços de tráfego reais. A partir dos parâmetros do modelo fuzzy TSK, derivamos uma expressão para a taxa da fonte controlável a qual minimiza a variância do tamanho de ?la no buffer. O controle de congestionamento proposto é então aplicado em diferentes cenários de rede com vários nós. Comparações realizadas com outros métodos de controle de congestionamento demonstram que o controle de congestionamento proposto obtém menores taxas de perdas e consegue de fato manter o tamanho da ?la no buffer abaixo do valor desejado / Abstract: The present work proposes a adaptive control of traf?c ?ows based in the TSK fuzzy model. In this control, the TSK fuzzy model is used to predict in a manner adaptive the buffer length in one output link. With the objective of dynamically adjust the parameters of the TSK fuzzy model, we proposed a adaptive training algorithm. In the ?rst stage of the proposed training algorithm, the parameters of the premise and consequent parts of the model are obtained. The second stage consists of a re?ning algorithm of the parameters based in descent gradient. The effectiveness of the proposed predictor is evaluated through comparison with other adaptive predictors by using real network traf?c traces. With the parameters of the TSK model, we derive an expression for the controllable source rate which minimizes the variance of the buffer length. The proposed congestion control is applied in different network sceneries with several nodes. Comparison made with others congestion control methods demonstrates that the proposed congestion control obtain lesser loss rate and gets in fact to keep the buffer length below of the reference level / Mestrado / Telecomunicações e Telemática / Mestre em Engenharia Elétrica