Wavelet and VMD enhanced traffic forecasting and scheduling method for edge cloud networks

Abstract

With the proliferation of Intelligent applications, more and more organizations migrate their applications from cloud to edge cloud network to reduce the latency of applications and alleviate the workload of cloud. When the network congestion occurs, network monitoring and detection system may disable, and edge cloud network will be vulnerable to be attacked. Traffic forecasting can identify the traffic patterns in advance, and dynamically allocate network resources to decrease latency of applications and avoid security risks caused by network congestion. Therefore, we propose a network scheduling framework WVNF (Wavelet VMD Based Network Flow Management) based on traffic prediction, which utilizes neural networks to forecast network traffic and deploys route strategies to optimize network scheduling. Specifically, in order to accurately forecast network traffic, we propose a neural network model, named TSWNet (Traffic Sequence Wavelet Network). TSWNet uses VMD (variational mode decomposition) to decompose time series and extract signal structure information on different time scales, and adopts wavelet transformation to extract the local and global features of the traffic sequence in the time and frequency domain. In addition, we model this traffic scheduling problem and propose a route strategy, which utilizes the result of TSWNet to find the best path. In extensive tests, TSWNet significantly outperformed existing models, reducing MSE and MAE by up to 48.8% and 27.8% respectively, demonstrating its effective traffic prediction and network scheduling capabilities.