Deep Learning Architecture for Processing Cyber-Physical Data in the Electric Grid

Abstract

Due to the increasing complexity of energy systems and consequent increase in attack vectors, protecting the power grid from unknown disturbances and attacks using special protection schemes is crucial. In this paper, we discuss the machine learning component of the HARMONIE special protection scheme which relies on a novel combination of graph neural networks and Transformer models to jointly process cyber (network) and physical data. Our approach shows promise in detecting cyber and physical disturbances and includes the capability to identify relevant portions of the input sequence that contribute to the model’s prediction. With this in place, the end goal of developing automated mitigation strategies is within reach.