STE-NTP: A Long-Short Period Aware Network Traffic Prediction Model

Abstract

As the scale of network rapidly expands, the density and complexity of network connections have reached unprecedented levels, increasing the complexity of network management. Software-Defined Networking (SDN) enables efficient network modeling techniques by providing a controller interface, thereby implementing Network Traffic Prediction (NTP) and directly controlling underlying network hardware. However, existing NTP methods face challenges in handling the highly nonlinear and frequently bursty characteristics of network traffic, particularly in capturing and analyzing the spatiotemporal features of the traffic. To address this issue, this paper proposes an innovative NTP model, STE-NTP:Time-Space Encoding Based Network Traffic Prediction Model. This model utilizes advanced spatial and temporal encoding techniques to comprehensively process both spatial and temporal information, thereby improving prediction accuracy and efficiency. Additionally, an LTST-Extraction Block(long-term and short-term Extraction Block) is designed to enhance the model’s ability to predict long-term and short-term events in network traffic data through Long-term and Short-term feature extraction techniques.To further validate the model’s performance, 50,000 time units covering 200 routing schemes were simulated on the NSFNET and Geant2 network topologies using OMNeT++. The proposed STE-NTP model was then compared against other advanced prediction models in both short-term and long-term forecasting tasks.The results demonstrate that proposed STE-NTP exhibits significant advantages across multiple key performance metrics. These experiments not only validate the effectiveness of the STE-NTP model in predicting complex network traffic but also highlight its potential value in practical applications.