| （英） |
Virtual network functions (VNFs) make 5G networks more feasible to the diverse and heterogeneous communication environments. To deal with dynamic network traffic, VNFs need to be scaled up/down according to traffic load. With the development of artificial intelligence, forecasting-based VNF scaling methods have attracted widespread attention from academia and industry due to the capability of adjusting VNF configuration in advance. However, due to the existence of emergencies (e.g. disruptions), it is difficult to collect enough historical traffic data in some areas to train high-precision intelligent network traffic prediction models. This issue seriously affects the reliability and availability of 5G networks. To solve this issue, we propose a novel few-shot learning (FSL)-driven network traffic prediction approach for 5G VNF scaling. First, an FSL-driven forecasting 5G VNF scaling architecture is established, where the whole design principle and workflow are designed. Second, we devise the FSL-enabled network traffic prediction method, including intelligent model structure, training, and interference design. Third, simulation results demonstrate the effectiveness of the FSL-driven traffic forecasting approach for 5G VNF scaling. |