Metodologia para classificação de faltas em redes de distribuição baseada em redes neurais artificiais

Abstract

The present thesis proposes an intelligent scheme for fault classification in distribution systems, based on artificial neural networks and a sensitivity factor, defined as Negative Sequence Factor (F2). The scheme, which focuses on single-phase faults, uses current signals measured at the substation from which the distribution feeder to be protected originates. The proposed scheme detects and identifies the type of fault, indicating cable disruption during a fault, assisting protection systems, minimizing and even avoiding the risk of accidents, fires, death of animals and people. The F2 factor was developed to improve the performance of the protection scheme, since its behavior is different for the fault types studied: series - in which there is cable disruption without ground contact, shunt - where there the cable touches the ground without disruption, series-shunt with cable fallen by the source side and series-shunt with cable fallen by the load side. The latter two are characterized by the cable disruption with subsequent fall to the ground. Computational simulations have been performed using ATP/MATLAB and Hardware-in-the-loop in real-time, with Real-Time Digital Simulator (RTDS). The results show that the proposed scheme is efficient for classifying fault types in distribution systems and can increase the safety of medium voltage distribution network protection