Identificación de causa raíz de fallas por descargas eléctricas en líneas de transmisión

Abstract

This research develops an algorithm based on supervised learning by applying the nearest neighbour classifier (k-nn) in order to automatically analyze, characterize and classify an electrical fault based on its root cause. This paper analyzes two kinds of the origin of electric faults: lightning and fire. Taking as a starting point a fully detailed analysis of voltage ad current waveforms and environmental considerations, patterns can be found for the characterization of the origin of electrical faults, these attributes are fault inception point, fault impedance, fault phases, voltage level and climate. In this research, these characteristics are obtained from the modelling and simulation in the ATPDraw software of the most representative transmission towers with the lines that register the most failures of the National Transmission System of Ecuador. Finally, the combination of these characteristics and parameters are used to train and test the knn algorithm. The successful test of the proposed methodology demonstrates its validity for the identification of different types of failure according to its root cause. Furthermore, the developed algorithm has been trained and tested using a set of real field data. The results show that the methodology is robust and covers a wide range of root causes of failure.