Efficient Task Offloading and Resource Allocation in an Intelligent UAV-MEC System

Fantacci, Romano (proxy) (contact); Benedetta, Picano

Abstract : Nowadays, the functional integration of the Digital Twin (DT) technology and Artificial Intelligence (AI) method- ologies has made possible to reliably predicting the evolution of many random processes in order to efficiently support control and optimization procedures. According to this trend, this paper considers the joint use of these two technologies in an AI- empowered DT framework for an unmanned aerial vehicle aided multiaccess edge computing (UAV-MEC) system. In particular, this allows defining an intelligent UAV-MEC system capable of notably improving the quality of the service offered and flexibility in its deployment. We assumed the UAV-MEC network object of study as composed of elementary service areas, where each elementary service area comprises a set of small base stations. In such a context, a set of DTs provide decisions to lower the congestion level of elementary service areas by exploiting dedicated UAVs with on-board processing capabilities. To achieve this objective, a viable framework utilizing a matching game approach is suggested for effectively managing task offloading, channel allocation, and the dynamic assignment of the UAV sup- port to congested service zones within a same area. Additionally, a potential DTs architecture is outlined, conceptualizing each DT as a collection of basic cyber entities. Furthermore, comprehensive simulation results have been carried out to validate the efficacy of the proposed UAV-MEC intelligent system, as indicated by metrics such as task completion delay and accuracy in congestion prediction.​

Index terms : Intelligence system , machine learning , mobile edge computing , unmanned aerial vehicle , digital twin