Research On Cellular Network Base Station Traffic Analysis And Prediction Based On Machine Learning

With the continuous development of mobile communication technologies,the application of mobile communication networks is increasing,social life is increasingly dependent on mobile communication,and the demand for cellular network traffic is also increasing.The cellular network generates a huge amount of data,which contains information about users,networks and even the information of social development.Mining this information can provide important reference for the further development of communication technology,the deployment of operators' corporate strategies,and the progress of the social industry.This thesis is selected in the enterprise research project,focusing on the analysis and prediction of cellular network base station traffic data.Because of the spatio-temporal correlation between different base station traffic sequences,this thesis uses the clustering algorithm in machine learning to perform cluster analysis on the base station to obtain the traffic variation characteristics of each type of base station.On this basis,the neural network in machine learning is utilized.The network algorithm performs base station traffic prediction.The main contents of this thesis are as follows:1)This thesis reviews the research on data mining of cellular networks,summarizes the characteristics of cellular network data and related research directions,and combs the current research methods of data mining methods and data mining methods applied in the field of cellular network traffic analysis and prediction.2)For the characteristics of spatio-temporal correlation between traffic sequences of different base stations,this thesis uses clustering algorithm to the cluster analysis of base stations.The trend information contained in the base station traffic sequence is incorporated into the distance metric of the clustering algorithm to obtain a new distance-metric distance TBD(Trend-Based Distance).Since the distance metric can represent the spatio-temporal correlation between sequences,a K-Trend clustering algorithm for base station traffic time series is proposed.The base station is clustered into different clusters according to the spatio-temporal correlation between traffic sequences,and combined with geographical locations.The information analyzes the traffic variation characteristics of the base stations of each cluster.The performance of the proposed algorithm is verified by the actual base station traffic data set.The results show that the proposed algorithm has better clustering performance than the traditional K-means algorithm.3)For the traffic prediction problem of single base station and global base station,this thesis proposes two base station traffic prediction algorithms based on Long-Short Time Memory(LSTM)neural network.The first one is the TBD-LSTM single-base station traffic prediction algorithm.When predicting the future traffic of a single base station,the algorithm calculates the space-time correlation between the base station and other base stations,and selects the base stations with higher correlation to calculate the traffic of these base stations.The value is used together with the traffic value of the base station to be predicted as the input of the prediction model,training the neural network,and finally obtaining the prediction model of the base station to be predicted;the second is the K-Trend-LSTM global base station traffic prediction algorithm,which is for the K-Trend aggregation each class cluster after the class trains a prediction model,and all the base station traffic sequences in the class cluster are used as input to the cluster prediction model.The prediction model of each cluster can predict the future traffic of all single base stations in the cluster,value.The proposed algorithm is applied to the future traffic prediction of the base station in the actual data set.The results show that the proposed two algorithms have higher prediction accuracy than the traditional LSTM algorithm.