REGULADOR NEURODIFUSO ARTMAP DE VOLTAJE DE UN GENERADOR SÍNCRONO

Abstract

In this work, a neuro-fuzzy controller based on the fuzzy ARTMAP architecture is designed and simulated for regulating the terminal voltage of a synchronous generator connected to an infinite bus. Such architecture is one of the neuro-fuzzy models built from the Adaptive Resonance Theory for generating multidimensional maps between analog input patterns by an incremental and supervised learning. To begin with, the fuzzy ARTMAP architecture is directly applied as voltage regulator for the under testing generator; this task is accomplished by means of modeling an existing controller and replacing it later. The results obtained with this technique demonstrate the ability of fuzzy ARTMAP architecture for modeling and enhancing the response of a particular system in different operation points. Later, the fuzzy ARTMAP architecture is used for tuning a Mamdani proportional integral (PI) fuzzy controller. The results obtained with this technique are satisfactory for different operation conditions. In order to verify the neuro-fuzzy controller performance in a wider manner, the infinite bus machine system is subject to several tests within different events simulations such as three-phase short circuit faults, load changes and reference changes. The obtained results are compared with those offered by the reference controller, which is a type ST1 excitation system. From this comparison it can be observed that the neuro-fuzzy controller provide a more efficient and reliable voltage regulation than the ST1 regulator.