Mining And Predicting Traffic Patterns In A Large-scale Base-Station Network

With the popularization of 4G wireless technology,the traffic of wireless base station continues to increase.As the infrastructure of wireless network,it is very important to analyze the static and dynamic characteristics of base station traffic and excavate the evolving mode of base station traffic for the formulation of operation scheme and reasonable configuration of parameters of base station.At present,the existing measurement and model work focuses on the rule of traffic change in short time granularity(such as minute,hour,day).So far,there is no accurate measurement and pattern mining results of traffic change in a long time scale(such as one year)for urban base stations.Therefore,based on the traffic data of a large Chinese wireless network operator in a large city's base station network,this paper observes and measures the static and dynamic characteristics of base station traffic of more than 7,000 base stations in monthly units for one year,and clusters,analyses and predicts the time-varying patterns of base station traffic.The contributions of this paper are as follows:(1)A new clustering method for base station traffic evolution pattern is proposed.The method is based on the ranking sequence of the monthly total traffic value of the base station in one year.The experimental results on our data set show that this method can describe the fluctuation of waveform in short time series(without periodicity),and get clustering results that are easier to understand than traditional methods.(2)Based on this clustering method,we conducted a large-scale clustering analysis on the traffic evolution modes of more than 7,000 base stations of operators,and obtained six typical base station traffic evolution modes.The most important one covers 38.6%of the base stations,which is characterized by an overall upward trend of traffic,peaking in November and falling to a low in February next year.Other modes include"Spring Festival returns home" and "Double 11”e-commerce shopping mode.Combining with the characteristics of the city,we explain the reasons for the formation of various modes.These findings provide useful guidance for operators to master the rules of traffic evolution of their base stations.(3)A base station mode prediction method based on the location and address semantics information of the base station is proposed to predict the traffic mode of a new base station.Because the initial information of the new base station is very little and the prediction is difficult,we innovatively introduce the base station semantic label information into the base station traffic mode prediction.The experimental results show that the F1-score of the prediction model is improved by 5%by adding base station word vector representation,and the prediction accuracy of the two models is higher.The analysis results of large-scale actual base station network traffic have important practical value for network operators.The long-term traffic change model,clustering algorithm and prediction algorithm proposed in this paper are universal and can be applied to similar application scenarios.They have important theoretical value and practical significance.Figure 20,table 8,reference 43.