Traffic Prediction And Its Application In Ultra-Dense Wireless Networks

As the key technology in 5G mobile communication network,Ultra-Dense Networks(UDN)improve the spectrum utilization efficiency,increase network capacity and solve the blind spot problem.Under the UDN,the software defined network(SDN)technology is used to flexibly allocate network resources according to the different traffic flow of cells,and thus further improves the network efficiency.Therefore,it is necessary to accurately predict the traffic flow of each cell.However,as the cell radius decreases,the statistics of service users declines significantly and the traffic distribution between different small cells is more uneven.Due to these problems,the traditional machine learning algorithms can not be applied directly in the prediction of ultra-dense network business.In view of this,we propose a MLT-Boosting prediction method and improve the accuracy of prediction.Then,we study the influence of the flight schedule on the traffic distribution and propose an EMIST traffic prediction based on the spatiotemporal influence factors.At last,network resource is dynamically managed according to the predicted traffic to significantly improve the network performance.The main work and contributions of this work is listed as follows:(1)For the potential adaptability problem of Boosting algorithm under UDN caused by heterogeneous dense cells deployment and the mobility of the users,we propose an enhanced Boosting algorithm based on multi-step label transformation.The proposed algorithm can fully mitigate the effects of data skewness and distribution statistics by introducing the phase skewness detection and label compression mechanism,and fully alleviate the effect of skewness on the algorithm system.As a result,it can adjust the residuals dynamically in the iterative process.The proposed algorithm is more flexible than the original algorithm and the fitting ability is improved to some extent.(2)Influenced by the flight schedule,the users' business distribution in Airport presents the characteristics of sudden fluctuation without historical regularity.Addressing this issue,the spatiotemporal influence factors innovatively is proposed to capture the spatiotemporal influence comes from different flights around the boarding gate region.Then,this paper build the boarding gate region model by using the spatiotemporal influence factors and the residual fitting mechanism.Afterwards,overall region model is constructed by from the periodicity,tendency,regional and other aspects of airport user business.At last,this thesis employs the ensemble method to combine the boarding gate region model and overall region model to makeup the EMIST framework.(3)A series of experiments are designed and proves that the proposed EMIST system architecture's good performance with robustness,superiority and the portability in different regions in the airport.After that,by employing SDN architecture as application background,this thesis compares the dynamic resource allocation scheme with the prediction results from EMIST system architecture and the traditional fixed network bandwidth performance in terms of resource utilization and system throughput,the results proves the guiding significance of the proposed architecture to communication system.