Mobile Base Station Traffic Prediction Based on Traffic Data Analysis

Abstract

The mobile base station is an important communication hub, which plays a very important role in the whole Internet. On the one hand, during peak traffic periods, a large number of base stations present problems of load exceeding capacity, making network speeds very slow even when signal conditions are good. On the other hand, the tidal phenomenon of base stations makes it possible for the number of users to drop significantly during certain hours, thus making the traffic load problem of base stations increasingly important. To address this problem, this paper proposes a GRU recurrent neural network based mobile communication base station traffic prediction model and improvements. Using the open-source sample dataset provided by the organising committee of the China Universities Big Data Challenge 2021, the study was carried out by conducting a smoothness test on the time series data, extracting three key indicators, namely the average number of subscribers in a cell, the PDCP traffic in a cell and the average number of activated subscribers, cutting the dataset, with 80% as the training set and 20% as the test set, and importing the GRU neural network prediction model implemented in Python, the results were unsatisfactory. Therefore, the model was improved to build a Bi-GRU-based base station traffic prediction model, and the prediction data was tested, and the error MAPE was less than 0.1, which was a good prediction. The improved base station traffic prediction solution is of great significance to the application of strategies related to the dormant energy saving of base stations based on traffic prediction.