Learning-Based QoE Optimization for Green Edge Computing Networks With Aerial Servers

This paper considers an aerial edge computing (AEC) paradigm, where unmanned aerial vehicles (UAVs) are deployed as edge servers to provide computing services to mobile devices (MDs) in remote and hard-to-reach regions. Although offloading compute-intensive tasks to edge servers can improve the quality of experience and reduce energy consumption of MDs, it poses challenges in efficient system management due to the limitation in energy and bandwidth resources of the servers. To address this issue, we propose a green AEC architecture where servers can harvest energy from renewable resources such as solar power. We formulate an optimization problem that aims to maximize the MDs’ long-term satisfaction while guaranteeing sustainable operation for the UAVs by controlling computation offloading and resource allocation decisions. We leverage the Lyapunov optimization theory to handle long-term energy constraints in the formulated problem and then develop a deep deterministic policy gradient (DDPG) algorithm to solve the problem considering a dynamic network environment. We also integrate prioritized experience replay and weighted importance sampling techniques into the DDPG algorithm to improve learning performance. Experimental results demonstrate that the proposed solution achieves high performance and adapts well to network variations.