Load Balancing Technology Based On Traffic Prediction In Ultra-dense Wireless Networks

As one of the key technologies for the fifth generation of mobile communications,ultra-dense networks effectively improve the spatial multiplexing rate of resources and thus enhance the network capacity by intensively deploying network infrastructure.However,as the number of infrastructure increases,the load in each cell changes faster than before,which leads to a dramatic increase in the overhead of the load balancing scheme.Meanwhile,The load balancing method based on triggering will lead to severe lag,and the speed of convergence of the load balancing scheme based on dynamic network status cannot match the change speed of network state.In addition,as the network coverage radius becomes smaller,the randomness of the number of users accessing each cell will increase,and the load variation in the local area becomes more severe.This makes the general prediction method directly used in ultra-dense network traffic prediction unable to obtain high accuracy.As a result,it is difficult to provide effective information for load balancing.In view of this,we first propose a multi-dimensional time series forecast method and improves the accuracy of prediction.Then,we propose a hierarchical intelligent load balancing and reduce the network overhead and convergence time of the load balancing.Our main work and contributions are listed as follows:First of all,aiming to resolve the problem of large fluctuations in ultra-dense network traffic,a multi-dimensional time series prediction method is proposed,which alleviates the problem of large fluctuations in traffic flow and improves the accuracy of prediction.First,starting from the traffic prediction itself,the statistical learning regression model and the convolutional neural network in the deep learning are used to predict.Then,based on the correlation between the number of users and the traffic with the stability of the number of users,the traffic flow is further predicted.Finally,we weightedly sum the above three results to obtain a complete traffic prediction,and thus improving the prediction accuracy.Secondly,for the current load balancing scheme,the overhead is large and the speed is slow.This paper proposes a prediction based load balancing scheme combining centralized and distributed integration,statistical learning and swarm intelligence,which effectively reduces the overhead of load balancing,increases the convergence speed of the algorithm,and adapts to the dynamically changing of network.First,the system uses the particle swarm algorithm to search out the load balancing solution that maximizes the historical resource utilization of the system during the leisure period,which is used as the training data of the real-time load balancing scheme.Then,based on the real-time traffic forecasting results and the training data obtained from the previous step,the regression model is trained and a real-time load balancing solution is obtained.Simulation results show that the proposed load balancing scheme achieves resource utilization that is not less than directly optimized from centralized optimization with faster speed and lower overhead.