Deep learning-based identification of false data injection attacks on modern smart grids

Abstract

With the rapid adoption of renewables within the conventional power grid, the need of real-time monitoring is inevitable. State estimation algorithms play a significant role in defining the current operating scenario of the grid. False data injection attack (FDIA) has posed a serious threat to such kind of estimation strategies as adopted by modern grid operators by injecting malicious data within the obtained measurements. Real-time detection of such class of attacks enhances grid resiliency along with ensuring a secured grid operation. This work presents a novel real-time FDIA identification scheme using a deep learning based state forecasting model followed with a novel intrusion detection technique using the error covariance matrix. The proposed deep learning architecture with its optimum class of hyper-parameters demonstrates a scalable, real-time, effective state forecasting approach with minimal error margin. The developed intrusion detection algorithm defined on the basis of the error covariance matrix furnishes an effective real-time attack detection scheme within the obtained measurements with high accuracy. The aforementioned propositions are validated on the standard IEEE 14-bus test bench.