Detecção de ilhamento em sistemas de distribuição com geração síncrona distribuída

Abstract

Nowadays due to the increase of renewable energy sources, Distributed Generations (DGs) assumed an important role in distribution and sub transmission systems. However, abnormal operating conditions of the distribution system may cause that the DG remains isolated from the main system and connected to part of distribution system feeding its local loads. This operating condition is known as DG islanding. There are many island detection techniques however they can fail in some situation, such as in cases of low power unbalance. This work aims to present three news passive island detection technique for synchronous Distributed Generation that has better performance than the traditional one. The first technique aims to attenuates the oscillations caused in the frequency by short circuits and load switching using Mathematical Morphology. The MM reduces frequency oscillations, and in this way it allows to reduce the thresholds. With smaller thresholds is possible the detection of small power mismatch and in a much smaller time. The second technique is based on Frequency Oscillation Estimation in order to distinguish the islanding from other events that may occur in distribution systems. This island detection algorithm uses a small time window to estimate the oscillation frequency, obtaining faster responses than the existing methods which use larger windows to estimate the damping and frequency of oscillation. The third method aims to recognize de frequency shape through a pattern recognition algorithm. The algorithm uses singular value decomposition and linear discrimination analysis to classify the events. The algorithms performances have been tested considering several generation and load scenarios including short circuits, load and capacitor switching, DG outage and islanding. The proposed methods were compared with the island detection method commonly used, i.e. Rate of Change of Frequency and Under / Overfrequency. The proposed methods are reliable since they do not trip for the most of non-islanding event. Among the tested methodologies, the method of the Frequency Oscillation Estimation has the best performance. Its detection time is less than 40ms and it Non Detection Zone is smallest than the range between + 3.2% and -1.6% of the GD rated power. The mathematical simplicity of EFO is adequate for practical relay implementation and the results show that EFO performs better than the other techniques.