REACT: Reinforcement learning and multi-objective optimization for task scheduling in ultra-dense edge networks

This paper addresses the challenges of task scheduling and resource allocation in ultra-dense edge cloud (UDEC) networks, which integrate micro and macro base stations with diverse user equipment in 5G environments. To optimize system performance, we propose REACT, a novel two-level scheduling framework leveraging reinforcement learning (RL) for energy-efficient task scheduling. At the upper level, RL-based adaptive optimization replaces conventional power allocation techniques, dynamically minimizing transmission energy consumption under the Non-Orthogonal Multiple Access (NOMA) protocol. At the lower level, the joint task offloading and resource allocation problem is modeled as a multi-objective optimization challenge. This is solved using a hybrid approach combining meta-heuristic algorithms and Long Short-Term Memory (LSTM) predictive models, maximizing response rates and system throughput. Sensitivity analyses explore the effects of user density, channel quality, workload, and request size on performance. Comparative evaluations against state-of-the-art methods demonstrate the proposed framework's superior efficiency in tackling dynamic scheduling challenges, achieving energy savings and enhancing user experience.