Traffic Prediction And Mobile Terminal Vertical Locating For Mobile Network Optimization

Virtualization technology make the core network resource allocation more flexible and efficient.This requires the network to predict traffic in advance.Through the traffic and the number of users to predict user behavior.According to the results of prediction,limited network resources are allocated to improve the quality of service.There are more and more high-rise buildings in the city.The depth of coverage,especially the vertical coverage of the optimization becomes more and more important.Test personnel can not enter the high-rise building is also the difficulty of this problem.Seeking an effective way to use user data to determine the quality of signal coverage in high-rise buildings is also blank in academic and industry.First,this paper introduces the network architecture of LTE core network and access network.The control plane signaling is given through the LTE system interface,and the process of obtaining the number of users,the traffic volume and the user-perceived Wi Fi information are analyzed by using the control plane signaling and the user plane data.The principle and characteristics of time series are studied,and the principle of ARIMA model and the standard of model identification are introduced.The criteria for the prediction and the evaluation criteria are given.Second,this paper first analyzes the data characteristics of hourly traffic,and gives the model of forecast traffic based on analysis.The application conditions of ARIMA model in the traffic forecast are studied,and the modeling steps such as model identification,order determination and residual test are given.The STL method is used to decompose the traffic time series into seasonal,trend and random components.The seasonal adjustment of the sequence is then fitted with the ETS model,and the predicted data is used as the result of the seasonal components.Holt-Winters additive model and multiplicative model are analyzed.The process of modeling and forecasting using BP neural network is discussed.In this paper,the four models are used to predict the traffic and terminal quantity of a certain operator in a certain area.Experiments show that the accuracy of prediction can meet the requirements of mobile network performance optimization.Models are implemented by code and applied to the operator's actual core network.Finally,this paper uses the LTE network user plane depth packet analysis to obtain the Wi Fi physical address and RSSI which can be perceived by the user in the specific area.Based on the Wi Fi information obtained from the high-rise buildings,the energy matrix of Wi Fi is established,and the elements of the matrix are the RSSI value of Wi Fi at sampling points.The correlation coefficient of the two Wi Fi is the correlation between the Wi Fi columns corresponding to the energy matrix,and then the correlation matrix of Wi Fi is obtained.By using four algorithms of K-means,PAM,spectral clustering and Fast Unfolding,the unsupervised clustering analysis of Wi Fi samples is carried out,and three Wi Fi clusters are obtained.After determining the underlying cluster,the correlation between clusters is calculated,and the height label of Wi Fi is obtained.Signal coverage quality of mobile user is determined according to the value of RSRP measured by the LTE terminal.The results show that methods have high precision and satisfy the actual wireless side network optimization requirements.