A fuzzy grey forecasting model for multi-scale internet traffic

The Undulating Grey Model (UGM) can be used to predict internet traffic, and it performs better than other Grey Models upon multi-scale traffics. However, UGM employs frequency spectrum of the traffic and needs to select the most significant frequencies as the parameters, which is hard to be decided manually. In this paper, we introduce clustering methods to find the significant frequencies. We find the best modeling length based on different clustering groups. We also find the value of the best modeling length is relevant to the frequency spectrum.