Open Access

ARTICLE

A Network Traffic Prediction Algorithm Based on Prophet-EALSTM-GPR

Guoqing Xu¹, Changsen Xia¹, Jun Qian¹, Guo Ran³, Zilong Jin^1,2,*

1 School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, 210044, China
2 Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology (CICAEET), Nanjing University of Information Science and Technology, Nanjing, 210044, China
3 Cyberspace Institute Advanced Technology, Guangzhou University, Guangzhou, 510006, China

* Corresponding Author: Zilong Jin. Email:

Journal on Internet of Things 2022, 4(2), 113-125. https://doi.org/10.32604/jiot.2022.036066

Received 25 November 2022; Accepted 31 December 2022; Issue published 28 March 2023

Abstract

Huge networks and increasing network traffic will consume more and more resources. It is critical to predict network traffic accurately and timely for network planning, and resource allocation, etc. In this paper, a combined network traffic prediction model is proposed, which is based on Prophet, evolutionary attention-based LSTM (EALSTM) network, and Gaussian process regression (GPR). According to the non-smooth, sudden, periodic, and long correlation characteristics of network traffic, the prediction procedure is divided into three steps to predict network traffic accurately. In the first step, the Prophet model decomposes network traffic data into periodic and non-periodic parts. The periodic term is predicted by the Prophet model for different granularity periods. In the second step, the non-periodic term is fed to an EALSTM network to extract the importance of the different features in the sequence and learn their long correlation, which effectively avoids the long-term dependence problem caused by long step length. Finally, GPR is used to predict the residual term to boost the predictability even further. Experimental results indicate that the proposed scheme is more applicable and can significantly improve prediction accuracy compared with traditional linear and nonlinear models.

---

Keywords

---