Webpage created under the purpose of a Master Dissertation entitled Deep Learning Applied to PMU Data in Power Systems. The work was developed in a partnership between FEUP – Faculdade de Engenharia da Universidade do Porto – and INESC TEC – Instituto de Engenharia de Sistemas e Computadores, Tecnologia e Ciência.

 

The analysis of power system disturbances is fundamental to ensure the reliability and security of the supply. In fact, capturing the sequence of system states over a disturbance is an increased value to understand its origin. Phasor Measurement Units (PMUs) have the ability to record these fast transients with high precision, by providing synchronized measurements at high sampling rates. Indeed, these events can occur in a few seconds, which hampers their detection by the traditional SCADA (Supervisory Control and Data Acquisition) systems and emphasizes the uniqueness of PMUs. With the advent ofWide Area Measurement Systems (WAMS) and the consequent deployment of such monitoring devices, control centers are being flooded with massive volumes of data. Therefore, transforming data into knowledge, preferably automatically, is an actual challenge for system operators.
Under abnormal operating conditions, the data collected from several PMUs scattered across the grid can shape a sort of a “movie” of the disturbance. The importance of WAMS is therefore sustained on their ability to capture the sequence of events resulting from a disturbance, helping the further analysis procedures. Driven by the amounts of data involved, this dissertation proposes the application of Deep Learning frameworks to perform automatic disturbance classification. In order to do so, a set of measurements from several PMUs – installed in the Low Voltage grid of an interconnected system – is used, from which representative patterns are extracted so as to endow a classifier of knowledge related to system disturbances.
In particular, the strategies herein adopted consist of the application of Multilayer Perceptrons, Deep Belief Networks and Convolutional Neural Networks, the latter having outperformed the others in terms of classification accuracy. Additionally, these architectures were implemented in both the CPU and the GPU to ascertain the resulting gains in speed.