Supervised Learning for Detecting Stealthy False Data Injection Attacks in the Smart Grid

Abstract

The largest and the most complex cyber-physical systems, the smart grids, are under constant threat of multi-faceted cyber-attacks. The state estimation (SE) is at the heart of a series of critical control processes in the power transmission system. The false data injection (FDI) attacks against the SE can severely disrupt the power systems operationally and economically. With knowledge of the system topology, a cyber-attacker can formulate and execute stealthy FDI attacks that are very difficult to detect. Statistical, physics-based, and more recently, data-driven machine learning-based approaches have been undertaken to detect the FDI attacks. In this chapter, we employ five supervised machine learning models to detect stealthy FDI attacks. We also use ensembles, where multiple classifiers are used and decisions by individual classifiers are further classified, to find out if ensembles give any better results. We also use feature selection method to reduce the number of features to investigate if it improves detection rate and speed up the testing process. We run experiments using simulated data from the standard IEEE 14-bus system. The simulation results show that the ensemble classifiers do not perform any better than the individual classifiers. However, feature reduction speeds up the training by manyfold without compromising the model performance.