Research On Resource Management And Optimization Of Mobile Cloud Networks For Vehicular Applications

With the development of society and technology,the number of vehicles is growing fast,causing great challenges over the transportation systems,environments and human health.The requirements of safe traveling,efficient transportation and good driving ex-perience become higher and higher.Meanwhile,the governors of transportation also need autonomous traffic management,control and congestion dispersion.These challenges have become global problems.Nowadays,vehicular cloud networks turn into an effective solu-tion to these problems.However,the high latency brought by cloud computing is hard to meet the real-time requirements of Internet of Vehicles.The development of edge comput-ing provides a promising and efficient solution for the problem of architecture and resource management in vehicular networks.How to build an architecture that can satisfy the needs of vehicular applications and typical scenarios,and how to manage resources in efficient ways,are key questions in the researches of vehicular networks.In this dissertation,aiming at the diverse requirements of quality of services of vehicular applications,we propose a layered cloud networks architecture,and based on the architecture,we study the resource manage-ment and optimization issues in different layers of the architecture.The main contributions of this dissertation are summarized as follows,(1)Design and analysis of layered cloud networks architecture for vehicular applications,In this dissertation,the types of vehicular applications and typical scenarios are sum-marized.Then,a layered cloud network architecture consists of vehicular cloud,edge cloud and remote cloud is proposed.The detailed design of each layer is discussed,as well as the applications it serves.The relations between the architecture and the fol-lowing researches are also explained.Finally,the advantages of layered cloud network architecture are discussed in terms of application,resource utility,latency,reliability and expansibility.The architecture provides basic idea for the following researches.(2)Research on computing resources allocation policy in vehicular cloud,On the basis of the architecture,this dissertation studies the computing resources al-location problem in vehicular cloud.In highway synchronized flow,the computing resources of vehicles inside vehicular cloud can be shared to other vehicles.These vehicles can also offload their computing tasks to other vehicles in order to reduce the latency of completing the task and enhance the efficiency of resources utility.During this process,the social ties could influence both the number of resources a vehicle can share and the transmission path of tasks.This dissertation studies the comput-ing resource allocation policy under restriction of social relations in vehicular cloud.The above problem is formulated and the expressions of probabilities of state tran-sitions are analyzed.After that,the model is solved by model-based dynamic pro-gramming algorithm and the resource allocation policy is retrieved.The performance of proposed policy under different social threshold,request arriving rate and average almount of computation is compared with baseline policy.Simulation results show that the proposed policy performs better than baseline policy on all nmetrics.(3)Research on caching resources allocation policy in edge cloud,On the basis of previous research,this dissertation studies the caching resources al-location problem in edge cloud.At intersections of urban area,the trajectories of vehicles can be predicted with high precisions,which gives the possibility of realiz-ing proactive caching.However,the trajectory predictions are not perfectly precise,prediction error could waste a lot of caching resources.On the other hand,the total number of caches are limited in edge cloud,it cannot offer integrated caching services for all vehicles.These issues pose challenges on caching resources allocation in edge cloud.In this dissertation,a vehicle tra.jectory prediction based caching resources al-location policy in edge cloud is proposed.The problem is formulated and analyzed at first.Then,the model is solved by using model-free tabular reinforcement learning algorithm.The concrete training process is given in details.After that,the long-short term memory networks is introduced to learn the behaviors of vehicles with real ve-hicle traveling data.Based on the prediction results,the performance of proposed policy under different prediction accuracies and total number of caches is verified.The simulation results show that the proposed learning policy performs better than other policies.(4)Joint allocation of communication and caching resources based on the vehicle-cloud cooperative caching in edge cloud.Based on the previous researches,we keep on studying the joint allocation of com-munication and caching resources under the scenario of cooperative caching between vehicle and edge cloud.In highway free flow,vehicles can be highly mobile.which makes the edge cloud hard to provide high quality caching services.By using vehi-cle trajectory predictions,the requests data can be cached in both the edge cloud and the vehicles in the opposite direction.In this way,the requesting vehicles can get data from both base stations and opposite vehicles simultaneously,which enhances the data fetching rate and quality of caching services.However,the bandwidths of vehicle-to-vehicle and vehicle-to-infrastructure communications are both limited,while the caching resources in edge cloud are also insufficient.How to allocate bandwidths and caches become a key problem.This dissertation proposed a joint allocation policy of both communication and caching resources.The problem is firstly modeled and analyzed.Because both the state space and action space are continuous,the function approximation and policy gradient are introduced,and neural networks are used to approximate policy and action-value function.After that,the model is solved by us-ing deep deterministic policy gradient algorithm,and the detailed training process is discussed.Finally,the performance of proposed policy is compared to several base-line policies in simulations.The result shows that the proposed policy can get higher rewards than other policies under different vehicle speeds,requesting data size and edge cloud's cache size.