Research On Cellular Network Traffic Prediction Based On Big-data

The number of wireless devices and the rapid growth of data transmission demand bring great challenges to the operation and optimization of cellular network.Full analysis of network data,in order to achieve accurate traffic prediction and comprehensive evaluation of network traffic growth potential,is of great significance to meet the needs of user data transmission and reasonable allocation of network resources.Based on the real cellular network management data,this thesis studies the trend of cellular network traffic from two aspects:short-term fluctuation analysis and real-time traffic growth potential evaluation.At first,aiming at the analysis of short-term fluctuation law of cellular traffic,this thesis proposes a cellular network traffic prediction architecture with time series fluctuation pattern clustering.The prediction architecture can extract the fluctuation patterns(baseline feature)from the traffic series of cellular cells and cluster them,so as to effectively reduce the interaction of time series with different fluctuation patterns in the training model,while retaining the residual components(residual feature).Furthermore,a deep learning model based on long short-term memory network is proposed to learn the baseline features in each cluster.At the same time,it is assumed that the residual characteristics of each sampling time in the cluster obey the normal distribution.In order to ensure the completeness of the whole architecture,we estimate the probability parameters of residual features.Experimental results show that the proposed scheme can effectively improve the prediction performance compared with the reference scheme.Secondly,predicting the wireless traffic behaviors and evaluating the growth potential of wireless communication networks can bring in valuable insights to guide the operation and optimization of wireless network towards higher network investment profits.Aiming at the real-time evaluation of cellular cell traffic growth potential,this thesis proposes a set of cellular cell traffic growth potential evaluation schemes based on machine learning.Based on the historical data of cellular network,the scheme evaluates the traffic growth potential in a multi-dimensional way.Specifically,a multi-dimensional mapping-based traffic growth potential evaluation model is established,which links the carefully selected key network behavior features with the traffic by exploiting the powerful machine learning system,eXtreme Gradient Boosting(XGBoost).Furthermore,an improved quantum particle swarm optimization algorithm is applied to search the upper limit of the traffic that the cell can carry according to the model,so as to provide guidance for network optimization deployment,improve the level of network traffic,release the growth potential of traffic.