Resource Allocation And Trajectory Optimization In Multi-UAV Assisted Vehicular Networks

Unmanned aerial vehicles(UAVs),as flexible and convenient edge servers,are widely used in edge computing networks.The optimization of internet of vehicle(Io V)edge computing is also a hot research topic in recent years.Therefore,some UAVs can make full use of its flexibility to collect vehicles’ information,forward vehicles’ tasks and provide edge computing services for vehicles with its limited computing resources in some special scenarios.For instance,some regions with high vehicle flow time period differences,areas where roadside units are damaged unexpectedly,and the environment that the edge servers’ capacity are insufficient.However,there are many limitations and other auxiliary facilities for a single UAV to serve ground vehicles.This paper studies the vehicular network architecture assisted by multiple UAVs,which is a threelayer network architecture composed of vehicles,UAVs and other edge servers.Vehicles need transmit their status information or tasks’ data to UAVs,those data will be solved by UAVs or be forwarded to other servers for auxiliary calculation after being received by UAVs.Meanwhile,considering the limited communication resources,computing resources and energy of UAVs,this paper studies the resource allocation,trajectory planning and energy optimization of UAVs,in order to make full use of limited resources,ensure the quality of service for vehicles and maximize the utility of each edge node.Firstly,resource allocation and trajectory optimization in multi-UAV assisted vehicular networks are studied in this paper.Channel resources are allocated according to the real-time state of vehicles’ task and the service state of UAVs in the environment.Part of tasks’ data is dynamically offloaded and forwarded to other edge servers for auxiliary calculation according to the current task redundancy of UAVs and the status of computing resources.The trajectory of the UAV is planned in real time to reduce the communication delay according to the position state of vehicles and the position state of UAVs.The trajectory movement,resource allocation and dynamic offloading actions of UAVs is a mixed integer nonlinear problem.In this paper,the integer and non-integer variables of the above problems are solved separately,and an extended multi-agent reinforcement learning algorithm is proposed,which combines multi-agent deep deterministic strategy iteration,resource allocation and clustering algorithm.Secondly,this paper continues to study the UAV trajectory planning and energy optimization in the UAV assisted vehicle network in the urban environment.In urban environment,the communication channel between UAVs and vehicles needs to consider the influence of line-ofsight communication and non-line-of-sight communication because of the shielding problem of buildings.Moreover,Based on the energy consumption model of fixed-wing UAV,this paper studies the energy optimization problem in the trajectory planning process of UAV.Multi-agent reinforcement learning algorithm is adopted to obtain the continuous acceleration strategy of UAV in three directions in three-dimensional space,so as to achieve energy optimization.In addition,the energy of the UAV is since the UAV limited and it is a passive device.Effective energy management can not only extend the service time of the UAV,but also optimize the node utility of the UAV as an edge server.Finally,some simulation experiments are carried out to verify the proposed solution.The advantages of the extended multi-agent reinforcement learning algorithm proposed in this paper are obtained in the multi-UAV joint cooperation scenario by comparing with the single-agent reinforcement learning algorithm.The effectiveness of the proposed scheme in resource allocation,trajectory optimization and energy management are verified by comparison with other benchmark experiments.