Network Traffic Prediction Based on Decomposition and Combination Model

Abstract

In this paper, a combination model based on complementary ensemble empirical mode decomposition (CEEMD) is proposed. First, CEEMD is applied to decompose original network traffic to generate high-frequency component, low-frequency component, and residual component. Then, the high-frequency components are modeled and predicted using bi-directional long short-term memory (BiLSTM). The low-frequency components and the residual component are modeled and predicted using autoregressive integrated moving average (ARIMA). Meanwhile, considering that the BiLSTM model is influenced by the hyperparameters, an Improved Bald Eagle Search (IBES) algorithm is proposed and applied to optimize three hyperparameters of BiLSTM, avoiding the blindness and subjectivity of manual selection of parameters. Finally, the prediction values of BiLSTM and ARIMA model are summed to obtain the final predicted value of network traffic. The comparisons with other models proved that the proposed network traffic prediction model is closer to the real data, with the optimal performance indicators, which is very suitable for high precision occasions.