Machine Learning based Anomaly Detection for Cybersecurity Monitoring of Critical Infrastructures

Abstract

Managing critical infrastructures requires to increasingly rely on Information and Communi- cation Technologies. The last past years showed an incredible increase in the sophistication of attacks. For this reason, it is necessary to develop new algorithms for monitoring these infrastructures. In this scenario, Machine Learning can represent a very useful ally. After a brief introduction on the issue of cybersecurity in Industrial Control Systems and an overview of the state of the art regarding Machine Learning based cybersecurity monitoring, the present work proposes three approaches that target different layers of the control network architecture. The first one focuses on covert channels based on the DNS protocol, which can be used to establish a command and control channel, allowing attackers to send malicious commands. The second one focuses on the field layer of electrical power systems, proposing a physics-based anomaly detection algorithm for Distributed Energy Resources. The third one proposed a first attempt to integrate physical and cyber security systems, in order to face complex threats. All these three approaches are supported by promising results, which gives hope to practical applications in the next future.