Elliptic Envelope Based Detection of Stealthy False Data Injection Attacks in Smart Grid Control Systems

Abstract

State estimation is an important process in power transmission systems. Stealthy false data injection attacks (SF-DIA) against state estimation may cause electricity theft, minor disturbances or even outages. Accurate and precise detection of these attacks are very important to prevent or minimize damages. In this paper, we propose an unsupervised learning based scheme to detect SFDIA on the state estimation. The scheme uses random forest classifier for dimensionality reduction and elliptic envelope for detecting these attacks as anomalies. We compare the performance of the elliptic envelope method with four other unsupervised methods. All five models are trained and then tested with a dataset from a simulated IEEE 14-bus system. The results demonstrate that the elliptic envelope based approach provides the best detection rate and least false alarm rate among these five unsupervised methods.