Análise de componentes independentes aplicada em distúrbios de qualidade de energia

Abstract

This dissertation presents the development of a technique for power quality analysis, considering the case of multiple disturbances. The purpose is to separate the disturbances that are occurring in sequence or at the same time, in voltage signals of the electric power systems. For this, is used the philosophy of blind source separation, with the assumption that the sources (the disturbances) that compose the data available are not known a priori, nor how they were mixed. It is used for this purpose the recent non-supervised technique called Independent Component Analysis for the decoupling between the different disturbances that are independent and occurring during the same interval, from a mixture model of the original disturbances. This separation of disturbances is crucial as a pre-processing step, since the existing techniques of location in time and classification have reduced performance when multiple disturbances occur. Two structures are presented to separate multiple disturbances, with regard to the number of mixtures (measured signals) available: the first is applied when are available more then one synchronized monitoring device placed in different points of the electric system, and the second is applied when there is only one available monitor. In the second case, it is proposed to use a filter bank to generate a sufficient number of mixtures required to the use of independent component analysis in the estimation of the disturbances involved in a single measured signal. The performance of the proposed technique was derived with synthetic data.