Traffic Prediction on Communication Network based on Spatial-Temporal Information

With the development of the communication and computer science technology, the traffic prediction of the communication network has attracted more and more interests from the scholars, meanwhile, it is also a significant problem in the real world. A good prediction result can monitor the diversification of the traffic volume and give an early alarm of the outlier. A key challenge of the traffic prediction in the communication network is that how to combine the spatial-temporal information together to make full use of the data. In this paper, we get two observations: (1) At the same timestamp, different square has different traffic volume, while at the same square, different timestamp also has different traffic volume. (2) There exists some periodicity in the traffic volume data along time. To address the challenges we mentioned before, we propose a novel Multi-Channel Spatial-Temporal framework (MCST) to model the spatial-temporal information. The three-channel CNN can mine the spatial information and enrich the temporal information, while the LSTM can model the temporal information. MCST can fuse the spatial-temporal information together to achieve the goal of giving a better prediction. Experiments on the public dataset of the communication network in Milan verify the effectiveness of the proposed model.