Detecting a Stealthy Attack in Distributed Control for Microgrids using Machine Learning Algorithms

With the increasing penetration of inverter-based distributed generators (DG) into low-voltage distribution microgrid systems, it is of great importance to guarantee their safe and reliable operations. These systems leverage communication networks to implement a distributed and cooperative control structure. However, the detection of stealthy attacks with a large impact and weak detection signals on such distributed control systems is rarely studied. In this paper, we address the problem of detecting a stealthy attack, named MaR, on the communication network of a microgrid while an attacker modifies the voltage measurement with the reference values. We collect datasets from a hardware platform modeled after a simplified microgrid and running the MaR attack performed with a Man-in-the-Middle (MitM) technique. We use the collected datasets to compare different attack detection algorithms based on multiple categories of machine learning algorithms. Our results show that the Random Forest algorithm outperforms the others to detect suspicious packets modified by a MitM attacker with an accuracy close to 97%.