Construction of subsystem determination of attacks in cyberphysical systems by neural network methods

Abstract

In this research work analyzes and compares existing methods for describing data from cyberphysical systems, methods for detecting network attacks targeting cyberphysical systems, analyzes fundamental approaches and solutions in the field of cyberphysical systems security, and makes recommendations for supplementing existing approaches using new algorithms. The considered application of the neuroevolutionary algorithm of NeuroEvolution of Augmenting Topology using a hypercube for the analysis of multivariate time series describing the state of cyberphysical systems in order to identify abnormal conditions. After the modification, the algorithm allows almost completely configuring the target neural network without user intervention according to the specified parameters, including additionally creating intermediate network layers that were previously unavailable in the primary version of the algorithm. The method is verified on the TON_IOT DATASETS dataset. The system topology is the structure of the Internet of Things. The data are relevant, verified and correct, which allows them to be used for analysis and assessment of the accuracy of the approach under consideration. The obtained overall accuracy, proximity of solutions, values of False Positive Rate and False Negative Rate indicate the lack of retraining of the model and the high reliability of this method for detecting attacks in cyberphysical systems