Modelagem e classificação de distúrbios de qualidade de energia em sistema de distribuição com geração distribuída

Abstract

The growing energy demand associated with the increase of non-linear loads in the Electric Power System and distributed generation have increased the incidence of Power Quality disturbances in recent years. These problems compromise the functioning of the equipment and even the safety of consumers. That said, identifying such events in the network to isolate their origin is an important and necessary task for the correct operation of the electrical system. In this way, several classification techniques have been proposed in the literature, with emphasis on techniques based on Machine Learning, which offer great generalization capacity and accuracy. Therefore, this work proposes the classification of four main power quality disturbances, namely, voltage drop, harmonic distortion, impulsive transient and oscillatory transient. To this end, the performance of the Decision Tree and Light Gradient Boosting algorithms, associated with feature extraction by the Multiresolution Analysis method of Wavelet, is compared, which has the advantage of extracting features both in the domain of time and frequency. The collected signals were extracted via simulation of the IEEE 13 Bus Feeder with addition of a photovoltaic solar generation system to add the characteristic of distributed generation to the grid. The signals are collected in two strategically chosen buses in order to observe the propagation of the disturbance along the network. The results obtained were satisfactory, with superior performance of the Light Gradient Boosting algorithm in all bars, with an accuracy greater than 98% for the closest bar, proving the great applicability of this technique in the classification of these disturbances.