Metodologia de identificação de sistemas dinâmicos multivariáveis baseada em modelo Neuro-fuzzy no espaço de estados com inferência evolutivo tipo-2 intervalar

Abstract

In this dissertation, a methodology for the identification of multivariable dynamic systems based on state-space neural-fuzzy model with evolving type-2 interval inference is proposed. An evolving learning algorithm for interval type-2 neural-fuzzy inference system is presented, where the combination of a fuzzy clustering method based on participatory learning and Extend Kalman filter is used for estimating the type-2 membership functions (shape and footprint of uncertainty). For estimating the consequent proposition, a fuzzy state-space identification algorithm is proposed, where the Markov parameters are recursively estimated, which are used to computing the state-space parameters incrementally. The efficiency and applicability of the proposed methodology are demonstrated through computational results (nonlinear SISO dynamic system, combined complex nonlinear dynamic system, time-varying nonlinear dynamic system) and experimental results (four-stage evaporator and helicopter with two degrees of freedom).