Um algoritmo genético para estimar matrizes de tráfego de fluxos origem-destino

Abstract

The accurate estimating of the traffic matrices origin-destination from the traffic measured in links of an IP network is a complex problem for which has not been found suitable solution. This paper applies genetic algorithms (GA) to the problem of estimating and compares the results with those obtained by three other techniques already applied: Linear Programming, Bayesian Inference and approximation for Expectation Maximization (EM) algorithm. Besides the use of GA, another contribution of this work is an initialization algorithm of parameters that decreases the computational effort and aids the convergence for expected result. The experiments show a better performance of the genetic algorithm in terms of the estimated error.