Research On Resource Scheduling Strategy In Vehicular Edge Computing

With the development of artificial intelligence(AI),computer vision,depth perception and sensing technologies,and so on,vehicles have gradually evolved from traditional travel tools into intelligent and connected vehicles(ICVs)with intelligent and interconnected comput-ing systems.However,the emerging vehicular applications(such as LiDAR,surveillance video,online applications,high-definition maps)have a high requirement for networking and computing capacity,which poses a great challenge to the existing vehicular communica-tion network.Although the introduction of cloud computing has eased some of the pressure on data processing,a long delay will also be produced.The edge computing architecture,which has the advantages of a flexible resource management and a rapid system response,has attracted much attention in recent years.Therefore,integrating the edge computing ar-chitecture to vehicle communication network thus forming the Vehicular Edge Computing(VEC)has become a new research direction.VEC can provide powerful communication,storage,networking,and communication capacity.However,in the scenario of VEC,large amount of data and application request would be generated from vehicles.Due to different request types and requirements,computation and storage capacity of edge devices are also different,and communication and computation resources are limited,how to efficiently schedule the resources of vehicular edge computing to meet the service requirements of different vehicles has become a research hotspot.To this end,the dissertation focus on the spectrum and link resource allocations for vehicles without task offloading requirements,and the joint resource allocation for vehicles with task offloading requirements.The main work and contributions of this dissertation are threefold:(1)To address the problem of limited spectrum resource in vehicular network,an immune clone-based spectrum resource allocation algorithm is proposed.Due to the difference type-s of both vehicle and request,a two-dimensional priority evaluation method is first pro-posed.The vehicle with a higher two-dimensional priority value will obtain more spec-trum resources,which improves the quality-of-experience(QoE)of vehicles.To improve the spectrum efficiency,a graph coloring model method is used to reuse the spectrum and avoid co-channel interference.And the spectrum resource allocation problem is modeled as a mixed integer programming problem.To solve the optimization problem effectively,an immune clone-based algorithm in artificial immune system is proposed.The proposed algorithm has a faster convergence than genetic algorithm,and can quickly obtain spectrum resource allocation solution,which reduces the delay and promotes the traffic safety.(2)To address the problem of unbalanced content distribution load of road side unit(R-SU),a fuzzy logic-base optimal content replica vehicles(CRVs)selection algorithm and an immune clone-based wireless link resource allocation algorithm are proposed.Firstly,a two-dimensional priority evaluation method is utilized to comprehensively evaluate the priority of vehicles and content.To reduce the content distribution load of RSU,some collaborative vehicles is utilized as CRVs,which can assist the RSU to distribute task processing result content.Secondly,an optimal CRV selection algorithm is proposed based on fuzzy logic by jointly considering the relative velocity factor,path similarity factor,and channel quality factor.For the selection of RSU or CRV,an immune clone based wireless link resource allocation algorithm is proposed to accomplish the content distribution efficiently.(3)To address the problem of offloading scheduling for VEC,a unified data scheduling ar-chitecture integrating communication,computation,cache and collaborative computing is proposed,and a deep Q-network(DQN)-based data scheduling strategy is proposed.First-ly,a unified architecture integrating communication,computation,cache and collaborative computing is proposed by take full advantage of the idle computing resources of collabora-tive vehicles.Considering the remaining lifetime of task data and the caching state of data queue,a multi-queue model on both vehicle side and RSU side is established.Also,the Markov Decision Process(MDP)is utilized to analyze the state transition of data queue.To obtain the optimal data scheduling strategy,a DQN-based deep reinforcement learning framework is established to learn the optimal actions,aiming at effectively reducing the loss of task data processing under delay constraint.