XTM: A Novel Transformer and LSTM-Based Model for Detection and Localization of Formally Verified FDI Attack in Smart Grid

Abstract

The modern smart grid (SG) is mainly a cyber-physical system (CPS), combining the traditional power system infrastructure with information technologies. SG is frequently threatened by cyber attacks such as False Data Injection (FDI), which manipulates the states of power systems by adding malicious data. To maintain a reliable and secure operation of the smart grid, it is crucial to detect FDI attacks in the system along with their exact location. The conventional Bad Data Detection (BDD) algorithm cannot detect such stealthy attacks. So, motivated by the most recent deep learning (DL) developments and data-driven solutions, a new transformer-based model named XTM is proposed to detect and identify the exact locations of data intrusions in real-time scenarios. XTM, which combines the transformer and long short-term memory (LSTM), is the first hybrid DL model that explores the performance of transformers in this particular research field. First, a new threshold selection scheme is introduced to detect the presence of FDI, replacing the need for conventional BDD. Then, the exact intrusion point of the attack is located using a multilabel classification approach. A formally verified constraints satisfaction-based attack vector model was used to manipulate the data set. In this work, considering the temporal nature of power system, both hourly and minutely sensor data are used to train and evaluate the proposed model in the IEEE-14 bus system, achieving a detection accuracy of almost 100%. The row accuracy (RACC) metric was also evaluated for the location detection module, with values of 92.99% and 99.99% for the hourly and minutely datasets, respectively. Moreover, the proposed technique was compared with other deep learning models as well, showing that the proposed model outperforms the state-of-the-art methods mentioned in the literature.