Traffic Matrix Prediction Based on Cross Aggregate GNN

Traffic Matrix (TM) prediction is defined as the problem of estimating future network traffic matrices based on historical network traffic data. It is widely applied in network planning, resource management, and network security. In this paper, we propose a CAGNN (Cross-aggregate GNN) model for predicting the traffic size of OD (Origin-Destination) pairs, and develop a TM prediction framework based on this model.Unlike traditional GNN models, our CAGNN model leverages the cross-aggregation of information from the network node graph and edge graph to obtain node representations and edge representations in the network. This cross-aggregation approach allows us to integrate different forms of traffic information in the network, providing a foundation for training downstream prediction models. We validate our framework on real traffic data from the Abilene network and compare our prediction performance with that of the traditional LSTM-RNN framework. The experimental results demonstrate that our model achieves superior TM prediction performance.