Traffic Prediction Technology Based On Deep Learning And Its Application In Communication

With the rapid development of urbanization,the population of large and medium-sized cities in China has increased year by year,the number of vehicles per capita has increased,and the situation of urban traffic congestion has become more and more serious.At the same time,mobile users,vehicle equipment,etc.,in transportation vehicles have great Mobile business needs,such as voice,video,data and so on.In urban traffic jams(such as crossroads,etc.),a sudden large number of dense mobile users will bring a strong communication load to the mobile cellular network base stations serving the cell,eventually resulting in dropout rate and access congestion of the cell users,the quality of community communications services(QoE)dropped sharply.In view of the network congestion caused by a large number of dense mobile users accessing a base station,an effective solution is to deploy a small cell in a trafficcongested area within the coverage of a macro cell mobile network to provide a heterogeneous wireless network service.However,compared with the general heterogeneous wireless network,the urban traffic scene has the problem of frequent switching due to the fast change of the communication terminal's moving speed,and a reasonable wireless resource management scheme becomes more important.At the same time,all types of communication terminals have the same moving speed,trajectory,and position information as the vehicle,so the wireless network status can be estimated by predicting the traffic flow and combining the corresponding mobile service traffic model.The traditional traffic flow prediction model is simple but inaccurate.It is not suitable for the increasingly intelligent traffic system in recent years.For this problem,the more commonly used method in recent years is the use of statistical learning,machine learning,or deep learning(artificial intelligence algorithms)for traffic flow prediction.In this paper,through the in-depth study and understanding of the current popular prediction methods,a encode-decode model is implemented based on the long-short term memory models in deep learning,and it is applied to short-term traffic flow in a two-dimensional circular planar area.At the same time,three kinds of short-term traffic flow models that are more popular today are implemented,including ARIMA,Xgboost,and SAE,and compared with the model of this paper.The experimental results on the real traffic flow dataset show that among the four models,the model proposed in this paper is more excellent.Then,based on the prediction results of short-term regional traffic flow,this paper proposes a network state prediction method based on the mobile traffic model and the location and speed of the vehicle's mobile users,and designs a corresponding wireless resource management solution.It includes macro base station user automatic unloading scheme for heterogeneous network considering micro base station load and interference,and ON/OFF scheme of roadside Small Cell at traffic jam.In this paper,traffic simulation software(Sumo)was used to simulate the traffic flow required for the experiment,and a simulation was performed using the network simulation(NS3)to verify the accuracy and feasibility of the proposed solution and to analyze the network performance after adopting this scheme,the simulation results show that the program can alleviate the network congestion to some extent,improve user service quality,while reducing system energy consumption,to achieve the purpose of green communication.