Long-term network traffic forecasting method based on deep learning

Abstract

The invention discloses a long-term network traffic prediction method based on deep learning, which includes: first obtaining the regional network traffic sequence, and counting the traffic value used in each moment; then preprocessing the traffic matrix sequence to obtain the input data of the Transformer model ; Secondly, establish a Transformer model, and use the Transformer model to adaptively extract temporal and spatial correlations for two-dimensional matrix data; finally, use an adaptive training mechanism for model training. The invention improves the accuracy of multi-step long-term forecasting of network flow, facilitates operators to plan future network resources in advance, and is beneficial to rational allocation of wireless resources.

Description

technical field [0001] The invention belongs to the technical field of wireless communication, and in particular relates to a long-term network traffic prediction method based on deep learning. Background technique [0002] In recent years, the fifth generation of mobile communication technology (5th Generation, 5G) has developed rapidly, representing the main direction of future network development, and will drive the society to gradually expand from the broadband interconnection between people to the Internet of everything, so as to become more It has profoundly affected the way of life and work of human society. The rapid development of mobile networks and the multiplication of traffic data continue to affect network performance and user experience, and pose new challenges for operators to rationally allocate base station resources and effectively ensure network stability and user experience. In order to meet the huge traffic demand, network operators and managers must s...

AI Technical Summary

Problems solved by technology

However, although the existing traffic forecasting models have achieved good results in short-term forecasting, they are not effective in long-term forecasting.

Due to the suddenness and randomness of the traffic sequence, it is difficult to capture its temporal and spatial correlation dynamically, and the existence of error accumulation makes the long-term prediction accuracy drop sharply with the increase of time.

Images