Combined Neural Network for Assessing the State of Computer Network Elements

Abstract

Monitoring the status of computer networks is an integral part of network administration processes. The number of nodes in modern networks is constantly increasing the topology is becoming more complicated. It is becoming increasingly difficult for a system administrator to timely identify and eliminate contingencies. Specialized intelligent support systems or specialized knowledge bases can help in this task. As a rule, the basic of these systems are artificial neural networks. The paper considers the structure of a combined neural network, focused on solving the problem of assessing the state of computer network elements. Three training methods are considered: the stochastic gradient descent, the adaptive learning rate method, and the adaptive inertia method. The results of an experimental evaluation of various options for implementing a combined neural network and its training methods are presented. The results showed a high accuracy of calculations, good adaptability and the possibility of application in a wide range of computer network configurations.