Computation Offloading And Resource Allocation For Internet Of Vehicles

With the explosive development of Io T and 5G technologies,emerging computingintensive applications,such as autonomous driving,augmented reality,and virus scanning,have gained widespread popularity.These applications not only consums large amounts of computing resources but also put forward higher requirements for latency,but the limited computing resources of vehicles cannot support the deployment of such applications.Mobile Edge Computing(MEC)places computing and storage resources at the edge of the network close to the vehicle,bringing new solutions to the above problems.The vehicle can migrate tasks to the MEC server for execution,which reduces the computing pressure of the vehicle,shortens the end-to-end delay of the application,and improves the user’s quality of experience(QoE).However,MEC has limited computing and storage resources compared to cloud computing.Therefore,in the computation offloading of MEC,it is necessary to coordinate the offloading between vehicles,and use the limited MEC resources to meet more computing requirements of vehicles and improve service quality.This brings new problems and challenges to formulating accurate and efficient computing offloading scheduling schemes in MEC.This thesis explores the optimal strategies for computation offloading scheduling and resource allocation for single-vehicle computation offloading scheduling and multi-vehicle computation offloading and resource allocation scenarios.The main contents of this thesis are as follows:(1)Computation offloading scheduling strategy based on deep reinforcement learning.This thesis studies the offloading scheduling problem of single-vehicle computing in a highly dynamic scenario,in which all the dynamic factors,such as task characteristics,wireless environment,and vehicle movement,are taken into account.An offloading scheduling algorithm based on Proximal Policy Optimization(PPO)is proposed,which jointly solves the offloading position and offloading time of the task to minimize the long-term costs.PPO can learn excellent offloading scheduling policy without prior knowledge about the environment.Simulation experiments show that the proposed algorithm can achieve better results than the baseline algorithm in different environments and user preferences.(2)Collaborative computation offloading and resource allocation strategy based on multi-agent deep reinforcement learning.The offloading decision and resource allocation strategy in the vehicular MEC environment are investigated.The constraints of computation and communication resource limitation,task deadline requirements,and the mobility of vehicles are taken into account.At the same time,a joint task type and vehicle speed perception delay model is modeled according to the task type and vehicle speed.To obtain a near-optimal solution of the formulated problem,The paper proposes a joint offloading and resource allocation based on Multi-Agent Proximal Policy Optimization(MAPPO)to solve this problem.MAPPO employs centralized training and distributed offloading execution decisions,thereby reducing the impact of system non-static.Simulation results show that the proposed algorithm can achieve excellent performance in terms of task completion delay,vehicle energy cost,and processing revenue.