Novo Método de Identificação Fuzzy para Sistemas MISO e suas Aplicações

Abstract

System identification techniques are essential for the knowledge of natural phenomena and processes from different sources. The main purpose of this research has been to build, theoretically and computationally, an approach to deal with data systems with multiple inputs and an output. Two distinct and well-defined steps are present in the identification process used in this study: data clustering and a Takagi-Sugeno-Kang type inference. The main contribution of this research is the construction of the antecedents and the consequences of the inference system in which new methods are used. One way to validate this new methodology has been through simulations with different databases. The experiments involved comparisons with other consolidated techniques such as: Nonlinear Auto-Regressive with eXogenous inputs, Hammerstein-Wiener and Multilayer-Perceptron (artificial neural networks). The coding has been done without the use of toolboxes, except for the tests, it has been done with the Compare routine of the Matlab software, which uses an accuracy measure called Normalized Root Mean Square (English). As a result of the comparisons, there has been a gain in precision in the response and a decrease in the computational effort. As an application of the new technique, data prediction tests have been carried out, with promising results, in addition to structuring a new methodology for the interpretation of signals coming from medical examinations, such as electroencephalograms. Future works include the introduction of the type-2 fuzzy set theory, an extension of the fuzzy set theory, to the inference dynamic with the objective of making the performance of the proposed methodology even better.