A Supervised Early Attack Detection Mechanism for Smart Grid Networks

The smart grid network is a sophisticated cyber-physical system with unique properties that provide new capabilities and challenges. One of the most significant challenges in deploying defensive strategies is the cascading failures induced by cyber-attacks. In this paper, we propose a novel attack detection mechanism based on supervised machine learning algorithms for smart grid networks. We propose a mechanism to generate a dataset using a realistic model for training a support vector machine and naive bayesian network algorithms. We generate a dataset using novel features to train the machine learning models by considering the heterogeneity of the smart grid network and interdependencies between the power and communication networks. We capture system parameters in each step of failure propagation to detect failures in the early stages of cascading failures before they percolate in the system. The proposed mechanism can detect cascading failures before they cause more damage and propagate through the system.