Title: Attention-based 3DTCN-LSTM short-term network traffic prediction model considering multi-base station spatiotemporal coupling

Authors: Yuliang Zhan; Ji Zhou; Jiayi Zhang

Addresses: Beijing University of Post and Telecommunications, School of Information and Communication Engineering, Beijing, 100000, China ' Beijing University of Post and Telecommunications, School of Modern Post (School of Automation), Beijing, 100000, China ' Beijing University of Post and Telecommunications, School of Modern Post (School of Automation), Beijing, 100000, China

Abstract: Implementing an accurate traffic prediction method can help telecom operators to pre-manage and optimise the network in advance, and it is also easy to adjust the power consumption of the base station. At present, the correlation between mobile devices and local base stations cannot be ignore, to accurately predict network traffic. Combining the spatiotemporal characteristics of traffic to achieve more accurate traffic prediction, this paper proposes a 3D temporal convolutional network-long short-term memory (3DTCN-LSTM) model optimised based on the attention mechanism. The mechanism reduces redundant information interference, enabling the extraction of long-range spatial correlations. The long-term dependency characteristics of the traffic are then obtained through the LSTM network. Finally, the experiments on the dataset demonstrate that the prediction effect of the 3DTCN-LSTM model is significantly better than LSTM, BiLSTM, TCN, TCN-LSTM, 3DCNN and other models.

Keywords: 3D temporal convolutional network-long short-term memory; 3DTCN-LSTM; attention model; network traffic prediction.

DOI: 10.1504/IJWET.2022.129253

International Journal of Web Engineering and Technology, 2022 Vol.17 No.4, pp.413 - 435

Received: 17 May 2022
Received in revised form: 20 Oct 2022
Accepted: 28 Nov 2022
Published online: 01 Mar 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article