5G Communication Network Optimization And System Research Based On Traffic Forecast

5G transmission networks need to evolve based on existing 4G transmission networks.During the evolution process,because of the upgrade of base stations and the growth of people's demand for the network,some nodes that are in transmission networks access too much traffic.This results in a high load on the access network and it also leads to an unbalanced load on the entire network.In this situation,changing the topological connection relationship between existing base stations and transferring nodes in a highload network to a low-load network is an effective way to improve the network load in a short period of time at a lower cost.For solving the problem of unbalanced load on the entire network,this thesis need to study topology optimization algorithms to fully explore the topology solution space.In order to ensure that the optimized topology will continue to be effective in future,the topology optimization algorithm cannot be based on historical data alone.Future traffic trends also need to be considered.That is why we need to study traffic prediction models.This thesis uses real telecommunication network topology data and traffic data to design a transmission network topology optimization system.This system based on spatio-temporal prediction,which uses historical network traffic and spatial information in the network topology to predict the future traffic of nodes.It also uses the predicted traffic and historical traffic as the basis for topology optimization,and uses genetic algorithms to explore the topology solution space to find a load-balanced network topology.The main contributions of this thesis are as follows:(1)This thesis proposes to use spatio-temporal prediction in transmission network,and considers that current spatio-temporal traffic prediction models lack effective method in measuring the time series similarity that has global similarity and local dissimilarity characteristics.Similarity distance that cannot effectively reflect similarity results in inaccurate traffic prediction.To solve this problem,this thesis innovatively uses a dynamic time warping method in the spatio-temporal traffic prediction model to calculate the similarity distance of two time series.Dynamic time warping automatically warps the time axis to calculate the minimum similarity distance and solves the problem,which means similar distance cannot correctly reflect the similarity caused by local dissimilarity.Experimental results show that the method improves the accuracy of traffic prediction by 6.7% over the traditional non-temporal prediction model.(2)Firstly,this thesis proposes a new genetic algorithm based on random node transfer to solve the problem,which traditional genetic algorithm violates the connection constraints between nodes due to the random connection of the edges.The new genetic algorithm based on random node transfer can fully search the solution space while the constraints are followed.Secondly,the set of equivalent nodes is constructed by hierarchically classifying nodes to reduce the dependency of inter-node,and it also assists genetic algorithm.Finally,a stack-based depth-first algorithm and a recursive deredundancy algorithm are designed to divide the whole topology into several subnetworks.Above algorithms provide the basis for the topology optimization algorithm.These methods finally achieve successful network optimization on a real large-scale network topology.The experimental results show that the genetic algorithm based on random node transfer improves the load balancing index by 8.4% compared with the heuristic node transfer algorithm.(3)By combining the above two methods,a real 5G transmission network topology optimization system based on spatio-temporal traffic prediction is successfully implemented.The experimental results show that the optimized network improves the load balancing index by 7.05% while carrying historical traffic and future predicted traffic.This proves the effectiveness of the system proposed in this thesis.26 figures,10 tables,41 reference articles.