Análise de padrões de descargas parciais por uma rede neural artificial.

Abstract

Insulators are a fundamental equipment for electrical power systems. They carry the responsibility of separating live parts from the neutral potential and avoid unwanted current circulation. When they fail, the repercussion of the problem could be extreme, carrying out to de-energizing lines and, ultimately cities, compromising the economy, public security and even the lives of the power grid users. There are three kinds of electrical insulators: porcelain, glass and polymeric. The ladder is in full expansion due to its reduced costs and good reliability. However, these equipments suffer the attacks of partial discharges when subjected to high electric fields. These discharges appear and disappear instantaneously in voids filled with air that are inadvertently in these equipments due to flaws of the construction process. The continuous action of these discharges leads to carbonization of the polymer, creating a favorable path for current circulation, culmination in insulation flaw and system breakdown. This work proposes the analysis of partial discharges patterns due to geometry, positioning and quantity of these insulation flaws, seeking to prove that the detection of this problem it is possible through the signals emitted by these discharges. For this, Artificial Neural Networks, which are computer algorithms for pattern recognition and classification, are used in problems that apparently don't have a simple relationship between the numerous variables involved.