无人机辅助的边缘计算空地动态协同机理与方法

In UAV-assisted edge computing, UAVs can be used as relays to help users at cell edge offload computation tasks to base stations, or play the role of computing platforms to execute computation tasks from users. Nevertheless, the cooperation of air-ground wireless-computation resource management (including scheduling, spectrum sharing/allocation, power allocation, computation task/resource allocation, etc.) and trajectory planning confronts new challenges: large-scale optimizations caused by complicated air-ground cooperation, three-dimensional complex coupled interference and unknown dynamics in dynamic environments. To handle these, our project first investigates the optimization for the cooperation of air-ground wireless-computation resource management and trajectory planning. The project will reveal the philosophy of the optimal cooperation. Afterwards, to address the problem of cooperation in dynamic environments with unknown dynamics, the project will exploit approximate dynamic programming to propose online-learning-based air-ground dynamic cooperation algorithms, in the presence of unknown dynamics. The project will reveal the philosophy of optimal dynamic decision for the cooperation of air-ground wireless-computation resource management and trajectory planning. Finally, in order to address the potential high computational complexity of the dynamic cooperation algorithms, the project will propose the method of fast air-ground dynamic cooperation, based on deep neural networks. Through the above research, the project is seeking breakthroughs in theory, mechanism and methodologies, so as to lay the foundations of low-latency low-energy-consumption air-ground edge computing.

无人机辅助的边缘计算中，无人机可作为中继协助小区边缘用户向基站卸载计算任务，也可作为计算平台执行来自用户的计算任务。然而，空地无线/计算资源管理（包含调度、频谱共享/分配、功率分配、计算任务/资源分配等）与航迹规划的协同面临新挑战：复杂空地协同引出的大规模优化问题、三维复杂耦合干扰、复杂动态环境中未知动态特性。为此，本项目首先研究空地无线/计算资源管理与航迹规划协同的优化机制，揭示最优协同机理。随后，针对包含未知动态特性的动态环境下的协同问题，本项目运用近似动态规划理论，提出基于在线学习的空地动态协协同算法，揭示面向空地无线/计算资源管理与航迹规划协同的最优动态决策机理。最后，本项目针对动态协协同算法的潜在高运算复杂度，基于深度神经网络，提出快速空地动态协同方法。通过以上研究，本项目寻求理论、机制、方法等方面的突破，为动态环境下的低时延低能耗空地边缘计算提供理论支撑和技术支持。

无人机辅助的边缘计算中，无人机可作为中继协助小区边缘用户向基站卸载计算任务，也可作为计算平台执行来自用户的计算任务。然而，空地无线/计算资源管理（包含调度、频谱共享/分配、功率分配、计算任务/资源分配等）与航迹规划的协同面临新挑战：复杂空地协同引出的大规模优化问题、三维复杂耦合干扰、复杂动态环境中未知动态特性。为此，本项目首先研究空地无线/计算资源管理与航迹规划协同的优化机制，揭示最优协同机理。随后，针对包含未知动态特性的动态环境下的协同问题，本项目运用近似动态规划理论，提出了基于在线学习的空地动态协协同算法，揭示了面向空地无线/计算资源管理与航迹规划协同的最优动态决策机理。最后，本项目针对动态协协同算法的潜在高运算复杂度，基于核函数深度神经网络，提出了快速空地动态协同方法。本项目成果在理论、机制、方法等方面为动态环境下的低时延低能耗空地边缘计算提供理论支撑和技术支持。