Research On Deep Learning-based Traffic Prediction And User Association In Heterogeneous Networks

With the development of mobile communication,smart mobile devices have been rapidly popularized,and mobile data traffic has also shown rapid growth.It is difficult for existing wireless communication technologies to meet this demand.To meet the above challenges,heterogeneous cellular network technology came into being.In heterogeneous cellular networks,although the dense deployment of large and small base stations increases the capacity of the system,the problems of network coverage and communication capacity in traditional networks have been solved to a certain extent.However,the cross-layer interference caused by the dense deployment of large and small base stations decreasing network performance also affects user experience.Based on the above background,this paper focuses on improving network performance,while takes into account user satisfaction,research and design of wireless traffic prediction and user association technology in heterogeneous cellular networks.The main research contents of this paper are as follows:1.Aiming at the time correlation of citywide wireless service traffic data,this paper proposes a citywide wireless service traffic prediction model based on densely connected gated circulation units.In order to fully mine the effective information in the unlimited business traffic data,the prediction model uses a sliding window mechanism to construct a citywide wireless business traffic data set.On this basis,through the design of densely connected gating cycle unit,with the help of the gating cycle unit has the feature of relatively long-term memory function for the input timing data,the time correlation of wireless service flow data is captured to improve the prediction of wireless flow Precision.Performance tests were conducted on actual data of various wireless services in Milan.The experimental results demonstrate that the proposed model can effectively capture the time correlation of wireless service traffic data,and significantly reduce the prediction error.2.Aiming at the problem that heterogeneous network users have insufficient understanding of the network environment during the association process,this paper proposes a user association method based on multi-agent deep reinforcement learning.Based on the establishment of a heterogeneous cellular network system model,combined with user satisfaction and communication operator cost constraints,through the intelligent association between users and base stations,a multi-agent Q-learning algorithm was designed to obtain the entire heterogeneous cellular network Maximize long-term utility.On this basis,considering the problem of large calculation amount of action space in the multi-agent Q-learning algorithm,a multi-agent deep Q network algorithm is proposed to learn the optimal association strategy.The simulation results show that this method can obtain the optimal user association strategy at a faster convergence speed through the continuous interaction with the network environment and the experience playback mechanism when the user has less known information about the network environment.