Estimação de estados do motor de indução utilizando rede neuro-fuzzy com aprendizado em tempo real

Abstract

This work treats the problem of state observation for the control of fast, non-linear dynamic systems, with disturbances, in real time. A new algorithm for state observation of dynamic systems based on a Neo-Fuzzy-Neuron with on-line training and convergence in just one step is propsed. It was necessary the use of a mapping equation between the observed states and the stades which are supposed to be compared with the measured ones. A theorem which allows the appopriate choice of a mapping equation was introduced. Expressions were developed for mapping equations with specific characteristics. These expressions can be extended to the obtainment of this kind of mapping equation. After that, the universe of mapping equations was restricted to a certain class. It was demonstrated an expression for the updating of the weights with optimum learning rate, which quarantees the observer convergence in just one step. Based on this proposal, four induction machine state observers were developed (one rotot fluxes observer and thrree rotor speed observers) whith mapping equations belonging to the referred class. These observers were appliend in an induction motor drive sytem. Induction machines are non-linear, multivariable and coupled dynamic systems, subjected to disturbances, whose control should be made in real time and whose time response should be fast enough for a satisfactory dynamic performance of the electric drive. These observers were validated through analyzes and simulations and experimental tests in the vector control of a squirrel-cage induction motor.