Prediction of corporate network traffic using artificial neural networks.

Abstract

The article investigates the forecasting of corporate network traffic with the use of artificial neural networks. A model that is able to analyze and predict Internet traffic over IP networks by comparing some training algorithms using statistical criteria is presented. The historical component of formation and development of complex systems is determined, and the concept of network traffic is offered. It is emphasized that neural networks are successfully used for modeling complex nonlinear systems and signal prediction for a wide range of engineering applications. They are one of the best alternatives for modeling and predicting traffic parameters, possibly because they can approximate almost any function regardless of the degree of nonlinearity and without prior knowledge of its functional form. A neural network model with the designation of the main inputs and outputs has been developed. The article proposes to use a multilayer perceptron based on the backpropagation method to predict corporate network traffic. The description of this method, its features and basic algorithms of training are given. It is emphasized that the inverse error propagation algorithm uses the gradients of the neuron activation functions to return the error measured at the output of the neural network and to calculate the gradients of the original error for each weight in the network. The mathematical component of each of the described algorithms of training of an artificial neural network with comparison in efficiency and productivity is offered. Methods for calculating the constant that distinguish different versions of the conjugate gradient are given. It is proposed to use this model using two Levenberg Marquardt algorithms and the elastic backpropagation algorithm to identify and manage corporate Internet traffic and as a fundamental tool for forecasting corporate Internet traffic at different intervals