Skip to Main Content
Rapid growth of computer network scales has made traffic matrix estimation essential in network management. It can be used in load balancing, traffic detecting and so on. Since traffic should be considered temporally and spatially, prediction is complicated. Tracking dynamic changes of traffic, reducing estimation errors and increasing robustness to noise are factors which should be considered in estimation. In this paper, we propose a novel method to estimate traffic matrix. This approach combines artificial neural network and evolutionary algorithms. It uses autoregressive model with exogenous inputs (ARX) joined with genetic algorithm (GA) which we call it ARXGEN. GA is used in gaining optimized weights and biases. To evaluate our method, we did our simulations on Abilene data. Results prove that it can well track dynamic nature of traffic and has lower estimation errors. It is also more robust to noise.