Cost Minimization Resource Allocation with Service Instance Caching and Task Migration for UAV Mobile Edge Computing

Air-ground integrated Mobile Edge Computing (MEC) is emerging as a promising technology to achieve the ITU 6G vision of ubiquitous connectivity. Thus, this paper integrates UAVs with ground base stations to provide connectivity for mobile users and to process their computation-intensive and delay-sensitive tasks. However, the limited storage capacity of UAV-side servers makes it infeasible to cache all types of service instances for various tasks. Additionally, the mismatch between server computing resources and user offloading demands leads to insufficient resource utilization, which ultimately deteriorates service delay. Meanwhile, the finite energy of UAVs also calls for reducing energy consumption. Therefore, we aim to minimize the cost, namely the weighted sum of total delay and energy consumption, through joint optimization of the UAV 3D hovering position, service instance caching, task migration, and computing resource allocation. Note that, this problem is characterized as a mixed-integer nonlinear programming (MINLP) issue with dynamic and uncertain system states. As such, we propose a two-stage approach to tackle it. Specifically, in the first stage, we develop an Asynchronous Advantage Actor-Critic (A3C)-based algorithm to design the service instance caching, task migration, and computing resource allocation, enabling multi-thread parallel environment interaction and policy training. In the second stage, UAV 3D hovering positions are optimized using successive convex approximation. Ultimately, we combine the two stages to obtain a high-quality solution. Simulation results demonstrate that our approach reduces the cost by 22.24% and 8.51% compared to DQN and DDPG while converging more quickly and stably.