Increasing demands to access the internet through mobile infrastructures have in turn increased demands for improved quality and speed in communication services. One possible solution to meet these demands is to use cellular traffic forecasting to improve network performance. In this paper, a model for predicting traffic at a selected cellular base station (BS) is proposed. In the model, spatiotemporal features from neighboring stations to the target BS are used, and this information is used for forecasting through a series of surfaces evolving over time and a deep learning architecture consisting of 3D convolutional networks. Experimental results showed that this method outperformed other approaches used to predict traffic data.

通过移动基础设施访问Internet的需求不断增加，反过来，对提高通信服务质量和速度的需求也越来越高。满足这些需求的一种可能的解决方案是使用蜂窝网络流量预测来提高网络性能。在本文中，提出了一种用于预测所选蜂窝基站（BS）流量的模型。在该模型中，使用了从相邻站点到目标BS的时空特征，该信息用于通过一系列随时间演变的表面以及由3D卷积网络组成的深度学习架构进行预测。实验结果表明，该方法优于其他用于预测交通数据的方法。