Research On Machine Learning With Performance Of Cooperative Communications In Vehicular Networks

Along with the progress of the research of intelligent transportation systems,the vehicular networks,as a typical application of the Internet of things in the field of transportation,play an increasingly important role from urban construction and development to people's travel efficiency.However,due to the high mobility of vehicles,traditional mobile computing faces challenges such as efficient and fast resource scheduling and power distribution.At the same time,the access network service between vehicles is one of the essential methods to provide communication services for near vehicles.Therefore,it is crucial to study the new architecture of the vehicular networks as close as possible to the communication deployments for the future intelligent transportation systems.Connectivity,as a fundamental and vital indicator of the vehicular networks,is of considerable significance to the network planning,topology control,and user experience of the vehicular networks.In recent years,experts and scholars have proposed many methods about the connectivity of the vehicular networks and achieved excellent results.However,due to the complexity of traffic environment,the characteristics that restrict the development of vehicular networks include complex wireless transmission environment,potential large-scale characteristics,high dynamic characteristics,partition network characteristics,network security,privacy,and other challenges.Therefore,on the basis of tracking the latest research progress and focusing on the performance indicators of the vehicular networks,this doctoral dissertation focuses on the in-depth research from the two aspects of cooperative communication and machine learning.Firstly,a cell-less cooperative communication network architecture based on mobile access points is proposed,and then a spatiotemporal distribution model of traffic in the mobile access points based on machine learning prediction is proposed.Finally,the precise resource scheduling and dynamic routing mechanism of mobile access points based on machine learning are presented.The core problem involves how to connect the vehicle terminals to the mobile networks to realize the dynamic,open,self-organized,easy-deployed,and low-cost networks.It includes the following research contents.In this dissertation,a cell-less cooperative communication architecture based on mobile access points is proposed.Compared to the traditional vehicular networks,according to different access modes of the vehicular networks,especially the horizontal access converged mode based on cooperative communication of mobile access points,by coordinating multi-point transmission and reception to communicate with cooperative base stations or mobile access points.The optimal access scheme of the vehicular networks can be realized to meet different requirements,establish a hybrid network architecture of the vehicular networks based on software-defined networks(SDNs)with low latency and high reliability.SDNs are used to collect vehicle motion state information to obtain the global network view,and the flexible scheduling of network resources is realized in the way of local first and then global.In the cellless vehicular networks,vehicles constitute multiple SDN clouds and realize the distributed and centralized resource scheduling and delivering mechanism.The simulation results show that the proposed converged cell-less communication network can save energy at both base stations and mobile terminals,and the energy efficiency of mobile terminals increases with the increase of the number of cooperative base stations,to optimize the performance of the vehicular networks.To solve the problem of frequent switching and interruption of high-speed vehicles,a 5G cell-less vehicular network communication scheme based on mobile access points is proposed.Fixed base stations are replaced by mobile access points to facilitate user access.Mobile access points use joint transmission and reception to cooperate with vehicle users to enhance the connectivity and reliability of vehicle-to-vehicle(V2V)communications.Three simple and feasible strategies for selecting vehicles as cooperative mobile access points are given to construct 5G cell-less mobile access networks.The data simulation results compare connection performance and latency performance under various mobile access point selection strategies.5G cell-less vehicular network communication scheme using mobile access points is significantly superior to simple mobile relay or mobile access scheme.Load balance is taken into account to enhance the connectivity and reliability of the vehicular networks.The second innovation point of this dissertation is to propose a spatiotemporal traffic distribution model based on machine learning prediction for the cell-less networks of vehicles,to model and analyze the traffic services in the vehicular networks.Aiming at the cooperative communication mechanism of V2V in the vehicular networks,a spatiotemporal distribution model with a cell-less structure based on machine learning prediction is proposed.According to the spatial and temporal distribution characteristics of data business requirements,a model is built for the spatial distribution characteristics of the vehicular networks based on the method of random geometry theory,a model is built for the temporal distribution characteristics of the business based on queuing theory,and a machine learning method is used to analyze and predict the traffic flow of the vehicular networks.It provides the basis for collaborative resource scheduling and distributed routing on the vehicular networks.This dissertation proposed V2V collaborative communication algorithm,integrates the access network selection mechanism of the terminal side and the regulation function strategy of the network side.According to optimal distribution and actual distribution of users,based on the objective function and constraint conditions of transformation,dynamic adaption to the change of the network adjustment function factors is given to guide the user terminals of the vehicular networks to reasonably select the dynamic nodes to access networks.A network resource allocation function based on QoE utility function is proposed to realize the optimal power distribution and spectrum resource sharing under cooperative communications.It solves the problem that different access modes of vehicle access networks will cause the loss and waste of some onboard resources.The third innovation point of this dissertation is to propose a precise resource scheduling and dynamic routing mechanism for space-time coordination of mobile access points based on machine learning.From the multi-level and multi-dimensional resource scheduling strategy,based on the business requirements of vehicles in the cell-less vehicular networks,with the time distribution,the space distribution,preception and projections of the business requirements of the vehicular networking,on different levels of large-scale macroscopic traffic flow model and small-scale microscopic traffic flow of urban dense traffic scene,and different dimensions in space and time,an accurate resource scheduling and dynamic routing prediction based on machine learning precision for space-time coordination of the mobile access points is proposed for dispatching and allocating the wireless communication resources in the vehicular networks.The resource supply and business demand are matched quickly to ensure the low delay of the vehicular network communication service.The multi-dimensional model combining space,time and frequency analyzes the performance indicators of the vehicular networks including connectivity under the precise resource scheduling strategy.The efficiency function of service quality is defined for key performance indicators such as time delay and connectivity of the vehicular networks.In the case of random interference,a discrete random approximation algorithm is adopted to optimize the scheduling parameters.Finally,in the scenario of the practical application of the vehicular networks,the relay nodes can be selected effectively to ensure the connectivity of the vehicular networks,according to the high connectivity networking algorithm rules given in this dissertation.So thereby,the priority of the users can be determined and resources in the frequency and time domain are allocated accordingly,for guaranteeing QoS requirements,wireless resource utilization,and user fairness.