Modelagem baseada em agrupamento nebuloso evolutivo de máxima verossimilhança aplicada a sistemas dinâmicos operando em ambiente não-estacionário

Abstract

This thesis presents a maximum likelihood based modeling approach applied to dynamic systems operating in non-stationary environment that uses recursive parametric estimation based on the method of fuzzy instrumental variable. The context is evolving and the idea is to guarantee a robust for estimation of the parameters of noise-corrupted experimental data. The methodology consists of an evolving fuzzy clustering algorithm based on the similarity of the data which employs an adaptive distance norm based on the maximum likelihood criterion that use an adaptive search strategy on the experiment in order to avoid the curse of dimensionality related to the number of rules created during data clustering of the data set. The computational and experimental results to exemplify the proposed methodology are: statistical analysis of the fuzzy instrumental variable inserted in the evolving context; black box modeling of a thermal plant; identification of a benchmark nonlinear system widely published in the literature and the black box modeling of a 2DOF helicopter. These examples are used to illustrate the performance and efficiency by operating in a non–stationary environment.