A Prediction Method of 5G Base Station Cell Traffic Based on Improved Transformer Model

In order to meet the network coverage and high quality, the proportion of 5G base stations in the global base stations increases year by year. The power consumption of the 5G base station is about 3 to 4 times that of the 4G base station, which makes the scale of the 5G base station grow rapidly. At the same time, the energy saving problem is increasingly concerned. By predicting the future traffic data of the 5G base station cell, the base station can be operated in advance to keep the energy consumption at a low level. Therefore, the accurate prediction of the traffic data of the base station is of great significance to the energy conservation and emission reduction of the current network. In this paper, we propose a time series prediction model for cell traffic data, which captures the coupling relationship between historical traffic data through the self-attention mechanism in Transformer model. Moreover, we add specific periodic term information to the positional encoding of Transformer model to make up for the lack of time sequence information in traditional Transformer model. The experimental results show that compared with the time series baseline model, the model has 15% and 8% respectively.