Characterization and Classification of Cyber Attacks in Smart Grids using Local Smoothness of Graph Signals

Abstract

Characterization and classification of cyber attacks in smart grids are crucial for situational awareness and mitigation of their effects. Graph signal processing (GSP) framework for the analysis of energy data, provides new perspectives and opportunities for such characterization by capturing topology, interconnections, and interactions among the components of smart grids. In this work, several forms of cyber stresses on power system's measurements and state estimation have been analyzed using the local smoothness of their graph signals. Using the local smoothness, characteristics of different cyber stresses are described analytically and evaluated by simulations. Moreover, the local smoothness features are used in machine learning models to classify multiple random and clustered cyber stresses and determine attack center and radius in case of clustered attacks.