Cyber-attack Detection on Distributed Frequency Control of Islanded MGs Using Machine Learning

Several coordinated control strategies along with extensive communication networks have been proposed and employed for proper monitoring and control of increased penetration of distributed generators (DGs) in microgrids (MGs). Because of communication networks, MGs are being exposed to numerous cyber-attacks with possibility of impacting active power sharing. In this context, an attacker can launch individual as well as coordinated data falsification attacks on distributed frequency control systems of MGs. This paper proposes a two-stage machine learning-based approach to detect and locate data falsification attacks on distributed frequency control of islanded MGs. In the first stage (regression), environmental parameters including solar irradiance, ambient temperature, and wind speed are provided to a machine learning regressor to predict active power from DGs. In the second stage (classification), a logistic regression compares the predicted active power of DGs with the measured active power of DGs to detect and locate data attacks in real time. Case studies on a modified version of the IEEE 123-node distribution system show that the proposed work can detect low margin attacks on distributed frequency control schemes with 97% accuracy.