Automated Real-Time Disturbance Report using Categorized Phasor Measurement Unit (PMU) Event Data
Author: Fannar Pálsson
Advisors: Samuel Perkin, Laurentiu Anton, Ragnar Kristjánsson, Birkir Heimisson
Published in: Skemman (Library of Research Publications in Iceland)
Accepted on: January 25, 2021
Abstract
Any uncertainty around a disturbance in the power system can be problematic for the operator when making decisions on how to proceed with the restoration process. Capturing the sequence of these disturbances, their uncertainties and how they affect the system state is improved by the use of Phasor Measurement Units (PMUs). These units provide synchronized measurements at the sampling rate of 50 Hz in the Icelandic power grid and are able to catch important characteristics of the disturbances. This thesis aims to find a procedure and a solution to classify these characteristics and to provide the operator with a detailed visual report in real-time of key parameters such as active- and reactive power, voltage and frequency, to explain the events and assist in further decision making. The thesis introduces the problem of classification for multivariate time series. To solve the problem, a convolutional neural network (CNN) was applied to the dataset. The networks outputs were then used to automatically generate detailed information and visuals for the assistance of the operators. The dataset consisted of 30 labeled events from the Icelandic power system, each containing 51 multivariate time series from the PMUs. The labels were four in total and consisted of a component trip, loss-of-load, islanding and oscillations events. The classification accuracy reached upwards of 94.7% and run-time results showed a great success of keeping it in real-time, for an event of 30 seconds, the report was able to be classified and visualized in under 10 seconds after the event. The results also presented the possible need for additional data to further improve the deep learning model and introduce opportunities for further research and future enhancements both to implementation and to a more detailed visual automatic report.
