| In recent years,with the development of wireless communication network technology,a large number of new wireless mobile services have emerged,such as augmented reality,image recognition,high-definition maps and online games.These services require wireless network communication and network computing and storage capabilities.Mobile Edge Computing(MEC)technology sinks computing processing to Edge Node(EN)and provides computing services by being close to user devices.However,for the Internet of Vehicles scenario,EN ’s limited computing resources often cannot fully meet the vehicle ’s offloading requirements within the delay constraint,and Vehicular Edge Computing(VEC)technology emerges as the times require.There are still many challenges in applying VEC to solve vehicle task offloading and resource management problems.Firstly,due to the high-speed mobility of the vehicle and the limited coverage of EN,the driving vehicle crosses multiple EN coverage areas in a short time,and frequent switching between ENs will cause problems such as decreased wireless communication quality and vehicle task interruption.Recently,researchers have proposed to interconnect EN to form an edge computing offloading network,and migrate the vehicle offloading task to the new EN for task calculation to meet the task requirements of the vehicle during driving.However,due to the uncertainty of vehicles driving in the road network,it is difficult to formulate an optimal offloading task migration strategy to control task migration costs and improve service quality.Moreover,due to the difference of vehicle computing requirements and the environment of EN,the migration of multi-vehicle offloading tasks in the road network is highly coupled between ENs,and how to realize the collaborative migration of distributed offloading tasks is facing challenges.In addition,the vehicle offloading demand in the road network shows the characteristics of time and space dynamic change,and the EN calculation ability of the corresponding section is relatively fixed and cannot be adapted to it.In view of the above challenges,the thesis focuses on the joint optimization of vehicle task offloading,vehicle task migration and heterogeneous resource management based on EN edge computing offloading network.The main research contents are divided into two parts:(1)Aiming at the influence of task migration calculation caused by the uncertainty of vehicle trajectory,the thesis applies machine learning technology to predict the vehicle’s time-related trajectory as the decision-making basis for the migration of offloading tasks between ENs.On this basis,the vehicle offloading task migration and heterogeneous resource allocation joint optimization strategy are further designed.The strategy takes into account the constraints of task migration objectives,channel bandwidth and computing resources.By adjusting the decision variables of vehicle offloading task migration and heterogeneous resource allocation,the joint optimization goal of balancing vehicle task processing delay and energy consumption is achieved.Aiming at the above optimization problems,the thesis designs a vehicle task migration decision and heterogeneous resource allocation algorithm based on Multi-agent Deep Deterministic Policy Gradient(MADDPG).The simulation results show that the proposed algorithm can reduce the average processing delay and energy consumption of vehicle tasks more effectively.(2)Aiming at the problem of uneven spatial and temporal distribution of vehicle computing offloading tasks on the road network,the thesis applies machine learning technology to predict the vehicle offloading demand of each road section in the road network,and uses Cloud Radio Access Network(C-RAN)technology to design the vehicle task offloading and heterogeneous resource management strategy under cloudedge collaboration.The strategy considers the constraints of cloud bandwidth pool,communication bandwidth,computational efficiency and task completion time.By adjusting the decision variables of C-RAN bandwidth allocation,communication bandwidth allocation and vehicle task offloading,the joint optimization goal of minimizing system energy consumption is achieved.In the thesis,the problem is further transformed into a Double Time-scale Markov Decision Process(DTSMDP).On the large time scale,the system predicts the traffic flow of each section,and schedules the bandwidth of EN through C-RAN technology.On the small time scale,the system allocates task transmission bandwidth and makes decisions on vehicle task offloading.In order to solve the optimization problem,the thesis designs a task offloading decision and heterogeneous resource management algorithm based on Deep Deterministic Policy Gradient(DDPG).The simulation results show that the proposed algorithm meets the requirements of vehicle task delay and reduces the energy consumption of the system. |
