ADAPTIVE NEUTRALIZATION OF CYBERPHYSICAL SYSTEMS STRUCTURAL BREACH BASE ON GRAPH ARTOFOCAL NEURAL NETWORKS
E. B. Aleksandrova, A. A. Shtyrkina Peter the Great St. Petersburg Polytechnic University
Annotation: The paper proposed a threat model of cyber-physical systems (CPS), with examples of attacks and consequences for systems for various purposes. It is concluded that the most critical consequences of attacks are related to the disruption of information exchange within the system. Thus, the task of ensuring the security of the CPS is reduced to restoring the efficiency of information exchange. To neutralize the negative consequences for information exchange, it is proposed to use graph artificial neural networks (ANNs). A review of modern architectures of graph ANNs has been carried out. To generate a synthetic training dataset, an algorithm was developed and implemented that simulates the intensity of the network flow and the workload of devices in the system based on graph centrality metrics. A graph ANN was trained for the task of reconfiguring the graph of the CFS network.
Keywords: cyberphysical systems, graph theory, spectral graph theory, graph artificial neural network.