Smart FDI Attack Design and Detection with Data Transmutation Framework for Smart Grids

Abstract

Conventional False Data Injection (FDI) attacks yield a distinct change of measurement values which can be easily detected using the state-of-the-art anomaly detection methods. However, if the attackers can learn the statistics of daily load measurement data (e.g., through snooping attacks), then smart FDI attacks can be designed to gradually alter the measurement characteristics over time to avoid detection. In this work, we provide a methodology to protect against smart FDI attacks. First, we create a smart FDI attack that can go undetected when using current state-of-the-art solutions. We then create a novel smart grid cyber defense framework that encrypts measurement data within the power grid and then decrypts data received at the control center to reveal attacked data samples. The proposed framework was validated using the IEEE 118-bus system. Performance was compared between the proposed framework, the Ensemble CorrDet with Adaptive Statistics (ECD-AS) bad data detection methodology and the quasi static model weighted least squares state estimator solution. Results show a mean F1-score of 95% for the proposed technique, 15% for ECD-AS and 18% for the quasi static least square method.