INTRUSION DETECTION IN INDUSTRIAL CONTROL SYSTEMS USING THE ENSEMBLE OF MODELS OF RECURRENT AND BIDIRECTIONAL GENERATIVE ADVERSARIAL NEURAL NETWORKS

Abstract

The paper considers generative adversarial and recurrent neural network architectures, as well as their application for intrusion detection in industrial control systems. For the experiments, the Secure Water Treatment dataset was used. This dataset describes the operation of the wastewater treatment plant. In the course of experimental studies, on examples corresponding to the normal state of the industrial process, recurrent and bidirectional generative-adversarial neuronal networks were trained. To improve the quality metrics, both networks were ensembled. The use of an ensemble of neural networks has improved the precision and recall.