Andrushchak V. Models for information and communication network flows management by using the methods of machine learning and artificial intelligence

The dissertation solves the scientific and practical problem of developing methods and models of information flow control in optical transport networks using machine learning algorithms and artificial intelligence, in conditions of high dynamics of change of probabilistic-temporal characteristics of information flows and conflicting requirements for service quality. A conceptual model of a software-defined optical transport network that provides the necessary infrastructure to support the developed intelligent algorithms for managing information and communication flows is proposed. This infrastructure provides and describes the rules of information collection for training, testing and deployment of appropriate models of intelligent algorithms for managing infocommunication flows. An algorithm for determining network states based on cluster methods of ML algorithms k-means and c-means has been developed. This algorithm allows you to build a sequence of events that allow you to predict with a certain probability of occurrence of a particular network event. This approach allows a more comprehensive approach to the management of infocommunication flows and take into account more network parameters. The method of payload aggregation at the boundary nodes of the optical transport network using deep neural networks has been improved. This approach has reduced the amount of workload with a small loss of packets while providing the necessary maintenance parameters. It is proved that reducing the amount of service information leads to a decrease in power consumption of the intermediate node by reducing the percentage of CPU usage of the node. The algorithm of intelligent control of infocommunication flows with the use of graph neural networks has been further developed. The developed algorithm, in contrast to the existing ones, allows to take into account the network energy consumption parameter as another FE element. A mathematical model for determining the energy efficiency parameter is presented. In the simulation process, it was proved that the developed algorithm for controlling infocommunication flows using graph neural networks made it possible to reduce the delay parameter during peak load hours by 18%. In particular, this model took into account the parameter of energy consumption, which was determined on the basis of its own methodology and in a separate simulation software. It is also proved by simulation that the developed algorithm of aggregation using a deep neural network allows to reduce the amount of service information by an average of 16%.