| The rapid development of mobile communication technology and artificial intelligence(AI)algorithms has prompted the metamorphosis of the existing transportation systems towards intelligent transportation systems(ITS).From the Internet to Internet of Things(Io T)to Internet of Vehicles(Io V),the process of the Internet of Everything affects every device around us gradually.From big data to recommendation systems and autonomous driving,the trend of the intelligent connection of everything changes our life and traveling gradually.As a model for the application of ITS that integrate the latest communication and computing technologies,the comprehensive upgrade of network connectivity and intelligence has made significant progress.However,there are still some problems in the current transportation system.The communication coverage and quality,computing resources and security cannot meet the growing user demand.The emergence of air-ground cooperative mechanisms provides a brand new opportunity to solve the above problems.The air-ground cooperative transportation system integrates aerial networks(satellites,airships,drones,etc.)providing communication and computing services in the air and terrestrial networks(V2V communications,V2 I communications,V2 N communications,etc.)based on cellular vehicle networking technology.The traditional transportation system is upgraded to a three-dimensional transportation network with full coverage of communication and computing services.However,the introducing of air-ground cooperation mechanism makes the new traffic system access more terminal devices,and the limited resources of multi-dimensional heterogeneity become one of the bottlenecks of development to meet the growing task demand of users.In addition,the rich data resources in the traffic system have been further expanded,and how to make good use of the traffic data for more efficient data processing has become an important issue to be solved.In order to solve the above-mentioned resource management and data processing problems,the key performance optimization of air-ground cooperative traffic network from the perspectives of intelligent framework,data collection,edge computing,and data prediction is the direction of this paper.Specifically,the main research contents and innovation points of this paper are as follows.1)An intelligent resource management architecture for air-ground collaboration and a UAV-assisted vehicle platooning resource management algorithm are proposed.First,a new intelligent resource management architecture is proposed,and the framework and components of the system are depicted in detail.Second,four major challenges of the intelligent resource management system based on air-ground cooperation mechanism are analyzed,including further improvement of communication,computation and caching efficiency,further reduction of end-to-end delay,overcoming the impact of vehicle mobility on resource management,and protecting the privacy of users in the transportation system.Finally,the potential opportunities of intelligent resource management system based on air-ground cooperation mechanism are sorted out,and AI-based optimization scheme design is carried out for the challenges and problems brought by the new communication technology to the future transportation system,respectively.2)A UAV-assisted resource allocation and trajectory optimization algorithm based on UAVs is proposed.First,for the ground sensor-based traffic flow data collection scenario,the UAV-assisted communication resource-constrained system is specifically analyzed,a joint communication and control optimization problem is proposed to maximize energy efficiency,and practical deployment-oriented constraints such as reflection coefficient,equipment scheduling,and trajectory optimization are considered.Second,considering that the problem under study is a hard non-convex problem,two deep reinforcement learning(DRL)-based intelligent decision algorithms are proposed,a Markov decision process-based problem reformulation is performed based on the original optimization problem,and the corresponding action space,state space,and reward function are designed to be applicable to the reinforcement learning(RL)solution.A deep Q-network-based RL algorithm is proposed to solve the original problem;and the inclusion of the design of adversarial neural network architecture is added to improve the convergence of the algorithm.Finally,the simulation results show that through the joint optimization of UAV flight trajectory,reflection coefficient,and equipment scheduling,the energy efficiency and fairness of the system can be improved The performance is improved by about 20% relative to the traditional non-intelligent algorithm,achieving sub-optimal performance,thus proving that the proposed DRL-based algorithm can give full play to the data-driven and large-scale complex problem processing of intelligent algorithms capabilities.3)A UAV-assisted resource allocation and computational offloading algorithm is proposed.First,an air-ground cooperative traffic system with a UAV swarm providing airborne mobile edge computing services is designed for the shortcomings of traditional mobile edge computing(MEC)systems,and a mixed-integer optimization problem is proposed to jointly optimize the offloading decision,computational resources and communication resources.Second,to effectively solve this problem,it is reconstructed as a MDP,and a centralized multi-agent RL algorithm based on DRL is proposed.Then,in order to solve the problem of centralized multi-agent RL algorithm,a distributed multi-agent federated RL algorithm based on RL and federation learning(FL)is proposed.Finally,simulation results show that the distributed algorithm is designed to avoid privacy leakage without sacrificing too much energy performance.Moreover,both intelligent algorithms based on DRL algorithms outperform the traditional non-intelligent algorithms in terms of total system energy consumption.4)A UAV-assisted traffic flow prediction and imputation algorithm based on UAVs is proposed.First,the problem of data deficiency in actual UAV collecting traffic flow data is addressed,and a comprehensive and practical analysis of the pattern of missing data is presented.Second,to improve the accuracy of traffic flow prediction under the condition of missing data,a novel deep learning(DL)-based traffic flow prediction and imputation model is proposed,so that the missing traffic flow data in the model can be filled in and still achieve high accuracy prediction.Finally,the simulation results show that the imputation method based on DL can obtain smaller prediction error than the traditional method when the missing data rate is less than 35%. |
