Deep Learning Applied to PMU Data in Power Systems

Abstract

With the advent of Wide Area Measurement Systems and the consequent proliferation of digital measurement devices such as PMUs, control centers are being flooded with growing amounts of data. Therefore, operators are craving for efficient techniques to digest the incoming data, enhancing grid operations by making use of knowledge extraction. Driven by the volumes of data involved, innovative methods in the field of Artificial Intelligence are emerging for harnessing information without declaring complex analytical models. In fact, learning to recognize patterns seems to be the answer to overcome the challenges imposed by processing the huge volumes of raw data involved in PMU-based WAMS. Hence, Deep Learning Frameworks are applied as computational learning techniques so as to extract features from electrical frequency records collected by the Brazillian Medfasee BT Project. More specifically, the work developed proposes a classifier of dynamic events such as generation loss, load shedding, etc., based on frequency change.