Detecting Anomalies in Cyber-Physical Systems Using Graph Neural Networks
Kristina V. Vasileva, D. S. Lavrova · Automatic Control and Computer Sciences · 2021
Abstract Application of convolutional graph neural networks for detecting anomalies in cyber-physical systems is proposed. The graph model reflecting the dynamics of variation in the state of devices is developed, and the algorithm for preprocessing the data providing the generation of the graph on the basis of the studied sample of telemetric indicator values is presented. Using experiments, the optimal parameters of the neural network are determined, the applicability and efficiency of the proposed model for revealing the anomalies in cyber-physical systems are shown, and the capability of the model to reveal and distinguish the classes of attacks is confirmed.