Anomaly detection in cyber-physical systems using graph neural networks

The paper proposes the application of convolutional graph neural networks to detect anomalies in cyber-physical systems, developed a graph model reflecting the dynamics of changes in the state of devices, presented an algorithm for data preprocessing, which provides the formation of the graph based on the studied sample of telemetry values. The optimal parameters of the neural network were established experimentally, the applicability and effectiveness of the proposed model for detecting anomalies in cyber-physical systems were shown, and the ability of the model to detect and distinguish between classes of attacks was confirmed