2022 VOLUME 2 ISSUE 2
RESEARCH ARTICLE
Traffic load predictions for cellular networks: A comparative study based on deep learning
AUTHOR
Miao Zhang, Chunshan Liu, Lou Zhao
ABSTRACT
Prediction of the traffic load of cellular networks is important for network planning, load balancing, and operational optimization. In this paper, a comparative study of cellular traffic load prediction models based on deep learning is performed and a prediction method built on a multi-channel Gated Recurrent Unit (GRU) model is proposed. The proposed method uses multiple channels to extract the daily and weekly variation feature as well as the variation feature of the peak period of the BS load and can be used to provide 24-hour ahead predictions. Experimental results obtained from real dataset show that the proposed multi-channel model can effectively capture the temporal-variations of BS load and reduce the prediction error. Compared with conventional prediction algorithms such as Convolutional Neural Network (CNN), Long Short-Term Memory network, GRU and combination of CNN and GRU (CNN-GRU), the proposed model can achieve better prediction accuracies.
KEYWORDS
Traffic prediction; 24-hour ahead prediction; GRU