Deep learning-based multilabel classification for locational detection of false data injection attack in smart grids

• Published in: Electrical engineering Vol. 104; no. 1; pp. 259 - 282
• Main Authors: Mukherjee, Debottam, Chakraborty, Samrat, Ghosh, Sandip
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2022; Springer Nature B.V
• Subjects: Algorithms; Deep learning; Economics and Management; Electrical Engineering; Electrical Machines and Networks; Energy Policy; Engineering; Machine learning; Monitoring; Original Paper; Power Electronics; Real time; Smart grid; Smart grid technology; State estimation; Unstructured data; Deep learning; False data injection attack; Cybersecurity; Power system security; Smart grid; State estimation
• ISSN: 0948-7921; 1432-0487
• DOI: 10.1007/s00202-021-01278-6

With the recent advancement in smart grid technology, real-time monitoring of grid is utmost essential. State estimation-based solutions provide a critical tool in monitoring and control of smart grids. Recently there has been an increased focus on false data injection attacks which can circumvent the traditional statistical bad data detection algorithm. Most of the research methodologies focus on the presence of FDIA in measurement set, whereas their exact locations remain unknown. To cater this issue, this paper proposes a deep learning architecture for detection of the exact locations of data intrusions in real-time. This deep learning model in association with traditional bad data detection algorithms is capable of detecting both structured as well as unstructured false data injection attacks. The deep learning architecture is not dependent on statistical assumptions of the measurements, it emphasizes on the inconsistency and co-occurrence dependency of potential attacks in measurement set, thus acting as a multilabel classifier. Such kind of architecture remains model free without any prior statistical assumptions. Extensive research work on IEEE test-bench shows that this scheme is capable of identifying the locations for intrusion under varying noise scenarios. Such kind of an approach shows potential results also in detection of presence of falsified data.