Hypergraph Partition And Deep Reinforcement Learning For Resource Allocation In 6G Vehicular Networks

In line with the communication development law of “commercial this generation,prestudy of the next generation”,China’s 5G technology has been widely commericalized in various industries,and “5G+Industrial Internet”gradually empowers the industry.As the most densely industrialized industry in modern times,the Internet of Vehicles is facing the future 6G vision and will expand the “5G+ Internet of Vehicles”with technologies such as artificial intelligence and edge computing.On the one hand,the Internet of Vehicles,as the communication infrastructure supporting the smart transportation system,is a powerful measure to solve the urban traffic dilemma.On the other hand,the optimization of the resource allocation of the Internet of Vehicles is directly related to green transportation,in line with the vision of “carbon peak and carbon neutral”.Firstly,in terms of network architecture and model,aiming at the shortcomings of existing single vehicle modeling,this paper designs a Poisson linear Co X point process for joint modeling of vehicles,roadside facilities and roads.Joint modeling can describe the high dynamic characteristics of vehicles in a more fine-grained manner.Secondly,for the highway car networking scene,combined with the correlation between vehicle-to-vehicle(V2V)communication and device-to-device(D2D)technology,the V2 V link reuses vehicle-to-vehicle infrastructure(Vehicle-to-Infrastrcture,V2I)link to improve the utilization of spectrum resources.For the derived problems of cofrequency interference and single link occupying multiple resource blocks,this paper proposes scheme of hypergraph partitioning into clusters and weighted 3-dimensional matching resource allocation.The simulation results show that the cluster design achieves the purpose of reducing interference within the cluster,and the clustering characteristics of the cluster are more flexible to maximize the reuse of V2 I resources.Finally,aiming at the scene of the Internet of Vehicles on urban streets,the timeliness of task offloading and the arrival rate of computing tasks are used as indexes to evaluate the efficiency of resource allocation of Internet of Vehicles.Considering the line-of-sight/non-line-of-sight model of urban streets,a task offloading method based on multi-agent deep reinforcement learning is proposed.The number of pre-matched nodes is determined by the coverage probability of offloading nodes.Furthermore,edge-cloud collaboration combined with linear Q function decomposition technology allows the agent’s decision record to be uploaded to the cloud as experience,and the cloud feeds back a more complete neural network model to the agent to assist in decision-making.The simulation results verify the effectiveness of computing task offloading from the perspective of power consumption and delay.The proposed scheme meets the needs of low latency and high reliability.