Statistics Department

Topic:

Machine Learning Techniques for Power Grid Protection and Control

Abstract:

Due to the diversity of mechanisms involved, it is challenging to protect and control power systems undergoing emergency operation such as cascading failures. Remedial action schemes (RAS) remain non-standardized and are often not uniformly implemented across system operators. The first part of this lecture will illustrate an open source power system dynamic simulator that integrates distributed and wide-area protective schemes. For example, load shedding and islanding have been successful protection measures in restraining propagation of contingencies and large cascading outages. We propose a novel, algorithmic approach to real-time selection of RAS policies to optimize the operation of the power network during and after a contingency. The algorithm is then tested with Monte-Carlo, time-domain simulations.

 The second part of this lecture focuses on the creation of an attack-resilient learning scheme for predicting the state of islanding or reconnecting microgrids. We build a classifier that uses machine learning techniques and Phasor Measurement Unit (PMU) data that is resilient to cyber-attacks. The goal of this learning scheme is to be able to determine dynamically whether the reconnection of an islanded microgrid would lead to a stable or unstable network. It is important that the process is robust due to the potential of PMUs being compromised during the decision to reconnect or not. The proposed machine learning algorithm makes use of a small set of secure PMUs to achieve relatively accurate predictions for the stability of reconnecting islands.

For more information about Eduardo Cotilla-Sanchez, click here.