Method for the Adaptive Neutralization of Structural Breaches in Cyber-Physical Systems Based on Graph Artificial Neural Networks

Abstract

This paper presents a model of threats in cyber-physical systems (CPSs) with examples of attacks and potential negative consequences for systems for various purposes. It is concluded that the critical consequences of attacks are associated with data exchange breaches within a system. Therefore, the CPS security problem is confined to restoring the data exchange efficiency. To neutralize the consequences, which are negative for data exchange, it is proposed to use graph artificial neural networks (ANNs). The contemporary architectures of graph ANNs are reviewed. An algorithm for the generation of a synthetic training dataset is developed and implemented to model the network traffic intensity and load of devices in a system based on graph centrality measures. A graph ANN is trained for the problem of reconfiguring the graph of a CPS.