Energy-Efficient Low-Altitude Caching-Offloading-Resource Allocation for Edge Intelligent Networks

Abstract

Low-altitude edge intelligence offers a flexible alternative to terrestrial networks for 6G wireless systems. However, joint optimisation of caching, offloading, and resource allocation remains challenging due to coupled delays, energy constraints, and dynamic content popularity. Existing single-agent or independent learning methods struggle with multi-unmanned aerial vehicle (UAV) cooperation and interference. In this letter, an energy-efficient multi-agent proximal policy optimisation algorithm is proposed for a three-layer UAV-assisted network, where task UAVs (TUAVs) aggregate users’ data, and service UAVs provide edge computation and cached files. Each TUAV acts as a decentralised agent, while a shared training module uses global system-state information to evaluate joint actions and guide policy learning. Simulation results show that our scheme reduces average delay and energy consumption compared to benchmark approaches.