Prediction of Cellular Network Channel Utilization Based on Graph Convolutional Networks

From the perspective of cellular network operators, cell-level prediction of channel utilization is crucial for proactive resource allocation, load balancing, and traffic control. One limitation of existing spatio-temporal prediction models is that the spatio-temporal correlation between cells and the rich properties of cells is not fully utilized. In this work, we propose a Graph Convolutional Network (GCN)-based architecture for the channel utilization prediction. Since CNN-based models are only suitable for capturing the spatial information dependencies when partitioning the regions into grids, we construct a graph where nodes represent cells and edges represent correlations of channel utilization between cells. A GCN and two gated linear units are then combined with learning spatial and temporal dependencies. In addition, more cell features, such as the type of cell coverage scenarios, are fed into the model to improve prediction accuracy. Experiments show that the proposed model outperforms baselines on a real-world dataset.