Intelligent Self-Optimization for Task Offloading in LEO-MEC-Assisted Energy-Harvesting-UAV Systems
Demeke Shumeye Lakew, Anh-Tien Tran, Nhu–Ngoc Dao, Sungrae Cho
Abstract
Given the notable surge in Internet of Things (IoT) devices, low Earth orbit (LEO) satellites and unmanned aerial vehicles (UAVs) have emerged as promising networking components to supplement the network capacity and ensure seamless coverage in 6G, especially over remote areas. However, task offloading and resource management are challenging to realize because of the limited connectivity duration of LEO satellites attributable to their high mobility and UAVs limited resources. Thus, this paper proposes a network model in which mobile edge computing (MEC)-enabled multiple LEO satellites in-orbit provide computational services for a resource-constrained energy harvesting UAV (EH-UAV). The EH-UAV collects data from remote IoT/sensor devices and periodically generates a computational task. To optimize the system model, we formulate a joint LEO-MEC server selection, transmission power allocation, and partial task offloading decision-making problem to maximize the service satisfaction and alleviate energy dissipation under the constraints of connectivity duration, task deadline, and available energy. To circumvent the non-convexity and dynamicity of the problem, it is reformulated as a reinforcement learning problem and solved using a novel mixed discrete-continuous control deep reinforcement learning ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$MDC^{2}-DRL$</tex-math></inline-formula> ) based algorithm with an action shaping function. Simulation results demonstrate that <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$MDC^{2}-DRL$</tex-math></inline-formula> effectively converges and outperforms the existing methods.