Energy-efficient Task Offloading for D2D-MEC via Multi-agent Deep Reinforcement Learning

The D2D-MEC technology enables the offloading of computation tasks to the edge server and nearby mobile devices, which has advantage in reducing the traffic burden of resource-constrained devices and improve the system’s energy efficiency. We study the task offloading problem for multiple user equipments (UEs) to minimize the energy consumption. We model the task offloading problem as a multi-agent deep reinforcement learning problem. To solve this problem, we propose a D3QN-based task offloading algorithm. Simulation results show that our proposed task offloading algorithm consumes lower energy compared with the baseline algorithms.