Spatial-Temporal Adaptive Convolutional Network with External Factors for Cellular Traffic Prediction

TL;DR: Wang et al. as discussed by the authors proposed an effective framework called spatial-temporal adaptive convolutional network based on external factors (STACN-EF) for cellular traffic prediction, which mainly includes five deep learning components: input spatialtemporal features division module, adaptive spatial graph convolution network (ASGCN), gate recurrent unit (GRU), history-future time conversion network (HFTCN), and gated fusion.

Abstract: Cellular traffic prediction is a prerequisite for the intelligent 5G network. However, accurate cellular traffic prediction is challenging. Because there are more complicated spatial and temporal correlations compared with road traffic prediction. And there are correlations between historical and future time steps in the long-term prediction. In this paper, we propose an effective framework called spatial-temporal adaptive convolutional network based on external factors (STACN-EF) for cellular traffic prediction. The framework mainly includes five deep learning components: input spatial-temporal features division module, adaptive spatial graph convolution network (ASGCN), gate recurrent unit (GRU), history-future time conversion network (HFTCN), and gated fusion. The input spatial-temporal features division module boosts the prediction's accuracy by learning the relationship between the central region and the hotspots region. Moreover, ASGCN and GRU are employed to extract inherent spatial-temporal correlations adaptively according to specific prediction tasks without any pre-defined information. In addition, HFTCN is designed to capture the relationship between prediction horizons and historical information and lead optimization direction, so as to improve the accuracy of long-term and short-term prediction concurrently. The experimental results in a real-world cellular traffic dataset show that STACN-EF outperforms state-of-the-art baselines and achieves optimal accuracy in all prediction horizons for the first time.