A Hybrid Deep Learning Method Based on CEEMDAN and Attention Mechanism for Network Traffic Prediction

Abstract

Accurate prediction of network traffic trends is important for self-management, intelligent scheduling and network resource optimization of base stations. Network traffic prediction is a prerequisite for intelligent scheduling of base stations, and accurate prediction will be beneficial for improving network utilization and energy saving in scheduling. In this paper, a hybrid deep learning method for network traffic prediction, CEEMDAN-TGA which consists of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Temporal Convolutional Network (TCN), Gated Recurrent Unit (GRU), and Attention Mechanism is proposed. Firstly, CEEMDAN is introduced to decompose the original network traffic data into different modes, then reconstruct the modes into trend sequence and noise sequence. Secondly, TCN is used to extract the short-term local features in the network traffic, GRU is used to obtain the long-term data-dependent features, and the attention mechanism is used to improve the prediction accuracy and stability. Finally, through the comparison of experiments, the prediction effect and accuracy of the proposed method are verified to have significant advantages, and the network traffic scheduling strategy is proposed on the basis of prediction.

Keywords

• Network traffic prediction
• deep learning
• complete ensemble empirical mode decomposition with adaptive noise
• temporal convolutional network
• gated recurrent unit
• attention mechanism