False Data Injection Attack Detection in State Estimation using Deep Learning

Abstract

Power grids are a complex system with a number of substations and transmission lines. Existing grids are converted into the smart grid due to the advent of digitization. For efficient control of the system, accurate state estimation is mandatory for which accurate readings of Phasor Measurement Unit becomes the base. State estimators are prone to False Data Injection(FDI) attack as it can pass through bad data detection mechanism. Covert cyber assaults framed by hackers, who have a deep understanding of power system, are dangerous as it cannot be detected by state estimators which results in the catastrophic effect. Thus an unsupervised attack detection algorithm is developed using autoencoder, which identifies the attack by examining the latest historical data and detect the state vectors which are not identical to the normal state vectors. This method is tested on IEEE 3 bus system with a wide range of attack plot.