Detection of Stealthy Cyber-attack in Distributed DC Microgrids Based on LSTM Neural Network
Xingquan Fu, Mengfei Niu, Guanghui Wen
Abstract
Distributed DC microgrids, which mainly consist of cyber layer and physical layer, can be classified as a classical cyber-physical system. Compared with the physical layer, the cyber layer has a significant risk of malicious cyber-attack due to its inherent vulnerability. This paper investigates a lightweight cyber-attack detection problem for distributed DC microgrids with bounded process and measurement noises. More precisely, the cyber-attack under consideration is assumed to be the stealthy cyber-attack. A long short-term memory (LSTM) neural network-based strategy is proposed to construct the attack detection scheme without using the information of the noises. Case studies indicate that the proposed attack detection strategy can work effectively in the following two cases: i) the statistical characteristics of system noises are unknown; ii) the magnitude of the cyber-attack is lightweight, which is detected extremely hard by traditional detectors such as <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\chi^{2}$</tex> detector.