Neural Network-Based Traffic Prediction Model with Adaptive Spatial-Temporal Analysis in NFV Networks

Network Function Virtualization (NFV) aims at satisfying various performance requirements of network services and increasing the performance/cost ratio mostly by decoupling software and hardware. However, network operators in the NFV scenario are facing several challenges in allocating resources and reducing operating expenses due to traffic explosion, and traffic prediction is helpful for operators to tackle this issue. The available models only consider the spatial or temporal correlation among traffic flows, but fail to combine the two aspects. In this paper, a neural network-based model with adaptive spatial-temporal analysis is proposed to predict the future traffic in NFV networks. First, we use the Long Short-Term Memory (LSTM) model to predict the traffic matrixes by capturing the long-term dependency. Then, we put forward a new mechanism to consider the spatial-temporal correlation between the target flow and the other flows in each traffic matrix. Besides, we adaptively select the most relevant flows, which are used to modify the prediction result through the Back Propagation Neural Network (BPNN). The experiments show that the proposed model can significantly improve the prediction accuracy compared to the available work.