Reinforcement Learning Based MEC Architecturewith Energy-Efficient Optimization for ARANs

Qiang He, Yingjie Lv, Li Zhen, Keping Yu

Abstract: Aerial Radio Access Networks (ARANs) are used to connect aerial nodes (such as satellites, aircraft, floating balloons) and ground infrastructures, which enables a global network coverage and provides a wide range of high-quality network services. At present, extensive researches are to integrate it with Mobile Edge Computing (MEC), to achieve more efficient data computing, data storage, and cache. In this paper, we primarily focus on exploring the edge computing architecture integrated with ARANs. Since the existing MEC architecture is not deeply integrated with ARANs, we propose the scenario of a complete four-tier MEC architecture that allows MECs and ARANs to collaborate effectively. Besides, for environmental protection and cost reduction, we propose a Q-learning algorithm based on the improved ϵ – greedy model to complete the MEC server selection and resource allocation. Finally, the simulation results are compared with other benchmark methods, and the effectiveness of the proposed method is proved. The energy consumption of the proposed method is significantly reduced.