Optical Network Traffic Prediction Based on Graph Convolutional Neural Networks

Understanding traffic patterns in largescale networks is of great importance for optical networks to implement intelligent management and adaptive adjustment. However, accurate traffic prediction in flexible optical networks is challenging because of the temporal and spatial autocorrelation of traffic. Spatial-temporal graph modeling is an effective approach to analyze the spatial relations and temporal trends of traffic in a system. We propose an efficient graph-based neural network named as the graph convolutional network with the gated recurrent unit (GCN-GRU). Based on a real-world optical networking traffic dataset, 98% accuracy for traffic prediction is achieved by GCN-GRU.