Research On Edge Computing Offloading Algorithm In Vehicular Networks Scenario

With the rapid development of the automotive industry and the Internet of Things technology,the increasingly abundant in-vehicle applications and services have caused great troubles for vehicles with limited computing resources.In response to the strict requirements of “low latency,high bandwidth,and high reliability” of the vehicular networks,introducing Mobile Edge Computing(MEC)technology is a good solution.It will sink cloud computing services to wireless networks on the edge,computing services can be provided near the vehicle terminal,which compensates for the delay and interference caused by the remote transmission process of the cloud server,and it is easier to meet the user’s service quality.Nowadays,academia and industry community have combined the vehicular networks with MEC technology,which has gained widespread attention.Due to the increasing number of delay-sensitive and high computational complexity invehicle applications,for resource-constrained in-vehicle terminals cannot met the above requirements.Therefore,this paper studies how to select offloading tasks,how to reasonably allocate communication and computing resources in the network,and minimize the total system overhead in the MEC-based vehicular networks.The main contents of this thesis are summarized as follows:(1)In light of the time-varying and complex vehicular networks,a semi-online computing task distribution and offloading algorithm based on competitive deep Q network(Dueling-DQN)is proposed.Due to the high complexity of the original optimization problem,we first build a vehicular edge computing(VEC)intelligent offloading system,and then decompose it into vehicle offloading action generation sub-problems and offloading behavior prediction sub-problems.By predicting different vehicle offloading behaviors,and calculating the cumulative reward value obtained after a series of vehicle offloading actions,the vehicle offloading decision can be effectively updated.The simulation results show that the algorithm improves the execution efficiency of computing tasks and load balance to a certain extent.Even the vehicle-mounted terminal with limited resources can better process a large number of data messages in real time and minimize energy consumption.(2)In view of the serious problems of increasing latency and energy consumption caused by the complex network status and huge data in the vehicular networks scenario,a Deep Reinforcement Learning(DRL)joint optimization algorithm based on Dueling-DQN is proposed,which solves the problem of traditional optimization algorithms falling into local optimality.Corresponding offloading strategies are made by improving the reward function,which minimizes energy consumption,calculation delays and communication costs.The simulation results prove that the proposed algorithm is more suitable for low-latency vehicular networks scenario in the real world.Compared with Reinforcement Learning(RL)algorithm and baseline algorithm,its performance is better and can effectively reduce the overall system overhead.