High impedance fault classification in microgrids using a transformer-based model with time series harmonic synchrophasors under data quality issues

Recent advances in distribution networks, driven by the integration of renewable energy sources, have spurred the emergence of microgrids, elevating concerns regarded reliability and stability. In this context, precise monitoring of events, particularly those elusive to detection like high-impedance faults (HIFs), becomes imperative. The development of phasor measurement units (PMUs) with their harmonic synchronized measurements has enhanced the monitoring task and fostering the application of synchrophasors even on microgrids. This work introduces a novel method for event classification in microgrids, utilizing combined low-rate PMU data and harmonic synchrophasor time series. Central to our approach is the usage of a state-of-the-art transformer neural network, based on the attention mechanism, to effectively discern HIFs from other faulty and non-fault events. Notably, this methodology accounts for prevalent PMU data quality issues, including noise, missing data, and synchronism errors. Results from real-world HIF data demonstrate a robust performance, with an accuracy rate of approximately 98% in event classification. This harmonic synchrophasor-based strategy showcases promise as an original approach for handling commercial PMU data, offering sufficient robustness for deployment in real-world applications.