Deep Learning-based Detection and Mitigation Strategy for Cyber-attacks on Advanced Metering Infrastructure

Given the direct connection of the distribution system with consumers, cyber-attacks are unavoidable. Therefore, ensuring the security of the distribution system is of utmost importance due to its critical role in maintaining the reliability and safety of energy delivery to end-users. As the Advanced Metering Infrastructure (AMI) collects data through various communication channels, there is an increased possibility of cyber-attacks. This paper proposes a Deep Learning (DL) based detection and mitigation strategy for attacks on AMI. The suggested algorithm is trained using a purity-based approach, which resolves the challenge of attacked data unavailability. Further, following the detection, a mitigation strategy is developed that can recover the lost/corrupted data. In addition to bill reduction theft attacks, the proposed algorithm has also considered many other real-time attacks on time-series measurements of smart meters with an attack detection rate of 98%. The proposed approach utilizes the sequence-to-sequence (seq-to-seq) long short-term memory (LSTM) based stacked autoencoders for capturing temporal dependencies in smart meter data which in turn may lead to an increase of about 12-38% in detection rate and a 4-45% in detection accuracy compared to benchmark detectors. Furthermore, the proposed strategy employs k-fold cross-validation to select an optimal threshold, which can reduce the false alarm rate by 2-6%.