Weighted fair energy transfer in a UAV network: A multi-agent deep reinforcement learning approach

Abstract

Flying Energy Sources (FESs) have been proven highly effective in transferring energy to battery-powered Unmanned Aerial Vehicles (UAVs). Such wireless transfer techniques in the literature were able to extend the flight duration of UAVs; however, they overlooked the fair distribution of energy among UAVs, which is of utmost importance for supporting diverse application demands. In this work, we quantify the urgency level of a UAV following its application responsibility, energy charging, and drainage rates and develop a framework for optimal energy transfer from FESs to UAVs in a weighted-fair way. Even though the developed framework can promote a highly balanced and fair energy distribution, its computation cannot always be done in polynomial time. Alternatively, to achieve a real-time solution, we further develop a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, namely AERIAL, to effectively model the complex interactions and dependencies among the UAVs and FESs and improve the wireless energy transfer process. The AERIAL system is implemented in the OpenAI Gym simulator platform and its performances have been compared with the state-of-the-art approaches. As high as improvements of 25.2% in fairness and 19.4% in average energy level demonstrated by the AERIAL system prove its effectiveness.