Lifelong Learning for AoI and Energy Tradeoff Optimization in Satellite-Airborne-Terrestrial Edge Computing Networks
Yuwei Wu, Beatriz Lorenzo, B. Liu · 2023
Satellite-airborne-terrestrial edge computing networks (SATECNs) emerge as a global solution for Internet of Things (IoT) applications in 6G. However, their highly dynamic nature with uncertain varying topology and network traffic makes their management and control more challenging. In this paper, we consider a scenario in which IoT devices, UAVs, and satellites with different edge computing capabilities make decisions to balance the freshness of information and energy consumption. Since SATECNs are highly dynamic and data freshness optimization requires timely decisions, we present a new lifelong learning computing resource allocation algorithm (LL-SATEC) that adapts to the environment by exploiting knowledge transfer between devices in different layers in SATECNs and previous experience. Lifelong learning is a promising machine learning algorithm that learns continuously without requiring a new training phase and avoids catastrophic forgetting. Our goal is to find computing resource allocation policies online for IoT devices, UAVs, and satellites that optimize the overall average age-of-information and energy trade-off. Numerical results show that our approach significantly accelerates learning compared to traditional reinforcement learning algorithms, achieves three times lower AoI and energy consumption in just a few iterations, and avoids catastrophic forgetting.