A new prediction method of alpha-stable processes for self-similar traffic

Because the self-similar processes have an infinite variance, the prediction method cannot be derived from the covariance. However, in the alpha-stable processes the covariation can be used to substitute the role of the covariance. Based on the theory of alpha-stable processes, a simple unbiased linear prediction method is developed for self-similar network traffic. The prediction method can be derived from the property of the covariation, and the prediction coefficients are solved from the cross-covariation matrix. The covariation orthogonality criterion ensures that the procedure of the prediction method is efficient and simple. The simulation experiments show that the new prediction method is able to predict the changes of the self-similar network traffic, especially in forecasting the bursty changes. As a result, this method can be used for network design so as to avoid network congestion.