: 10.62796/pijst
Data-Driven Semi-Supervised Approaches for Event Identification in Large-Scale Power Systems
Anuradha Krishna,
Assistant Secretary, SBTE, Department of Science, Technology and Technical
Education (Bihar)
Published Date: 02-01-2025 Issue: Vol. 2 No. 1 (2025): January 2025 Published Paper PDF: Download
Abstract- In this paper, we investigate three traditional SSL techniques for event detection: self-training, transductive support vector machines (TSVM), and graph-based label spreading (LS). Load loss, generation loss, line trip, and bus fault are four important event categories that may be classified using features extracted from synthetic PMU data using modal analysis. By comparing various methods on the South Carolina 500-Bus synthetic network, we find that graph-based LS is the most effective, demonstrating the usefulness of data-driven SSL techniques for detecting events in large-scale power systems. For real time monitoring and analysis, data-driven methods are becoming crucial due to the growing integration of Phasor Measurement Units (PMUs) and developments in data science. Unfortunately, fully supervised learning methods aren’t very effective because it’s hard to get enough labeled data, which is a problem because some grid events are rare and unclear. Because of its ability to use both labelled and unlabeled data to enhance performance, semi-supervised learning (SSL) becomes a potent option. Keywords: Semi-supervised, Event, Identification, Power, Detection.
Published Date: 02-01-2025 Issue: Vol. 2 No. 1 (2025): January 2025 Published Paper PDF: Download
Abstract- In this paper, we investigate three traditional SSL techniques for event detection: self-training, transductive support vector machines (TSVM), and graph-based label spreading (LS). Load loss, generation loss, line trip, and bus fault are four important event categories that may be classified using features extracted from synthetic PMU data using modal analysis. By comparing various methods on the South Carolina 500-Bus synthetic network, we find that graph-based LS is the most effective, demonstrating the usefulness of data-driven SSL techniques for detecting events in large-scale power systems. For real time monitoring and analysis, data-driven methods are becoming crucial due to the growing integration of Phasor Measurement Units (PMUs) and developments in data science. Unfortunately, fully supervised learning methods aren’t very effective because it’s hard to get enough labeled data, which is a problem because some grid events are rare and unclear. Because of its ability to use both labelled and unlabeled data to enhance performance, semi-supervised learning (SSL) becomes a potent option. Keywords: Semi-supervised, Event, Identification, Power, Detection.
