A Novel Deep Learning Framework to Identify False Data Injection Attack in Power Sector

Abstract

Monitoring the power sector in real-Time holds a paramount importance to modern grid operators. State estimation algorithms (SEAs) play a pivotal role in this domain by accurately furnishing the operating states of the grid. Lately a new genre of cyber attacks compromising the data integrity of the grid namely the false data injection attack (FDIA) has illustrated its detrimental consequences by efficiently evading the traditional bad data detection algorithm. Grid operators must ensure an efficient identification of such class of attacks residing in the measurement set, hence confirming a safe and reliable operation of the power sector. Identification of FDIA subsuming state forecasting policies can accurately capture the variations introduced in the set of estimated states due to an attack. This work advocates the implementation of a scalable neural network structure which showcases an effective forecasting policy of operational states, hence showcasing its potential to detect the presence of an attack in real-Time. The recommended neural network structure with its optimum hyper-parameters promote a superior state forecasting stratagem with the least error in comparison to most of the modern state forecasting approaches. A sedulous investigation on the IEEE 14 bus system effectually advocates the precursory postulates. © 2021 IEEE.