Abstract: Complexity and diversity of Internet traffic are constantly growing. Networking researchers become aware of the need to constantly monitor and reevaluate their assumptions in order to ensure that the conceptual models correctly represent reality. Internet traffic today is a complex nonlinear combination of the seasonal time series. The current network traffic measurement research is mainly concentrated on the flow forecasts and analysis based on network layer or transport layer. However, a single ARMA (n, n-1) model is used, which can only describe the overall network traffic trends, while different traffics based on the application layer aren’t always consistent with ARMA (n, n-1) model. Presented in this paper are traffic prediction models based on application layer, which use ARIMA seasonal multiple model (p, d, q)(P, D, Q)s for modeling and forecasting the seasonal time series from China’s exports of a metro area network link. Experimental results show that different application layer traffics perform different traffic behavior characteristics, and with the establishment of different application-layer flow prediction models, forecasting trends are very similar with the actual flow curves, and mean absolute percentage errors are around 10%. The authors firstly presents ARIMA seasonal multiple model as traffic prediction models based on application layer.