Intelligent Load-Aware gNB Optimization for Energy Saving in 5G via NWDAF-Orchestrated Framework

초록

Fifth-generation NG-RANs must jointly reduce energy consumption and preserve load stability under highly dynamic traffic and mobility, where static sleep-wake rules and reactive load balancing are inadequate. This paper introduces the Intelligent Multi-Agent RAN Controller (IMARC), a NWDAF-orchestrated framework that integrates multi-domain network analytics with feasibility-aware reinforcement learning control. IMARC coordinates two specialized agents: a Sleeper Agent that performs load-and stability-aware admission of gNB sleep actions, and an Energy Analyzer Agent that optimizes power states and sleep timers within admissible bounds, with all actions enforced through OAM to ensure stable execution. System-level simulations under non-stationary and bursty traffic show that IMARC reduces average power consumption from 726 W to 509 W, achieving a 29.9% energy saving, significantly outperforming RL-based NWDAF approaches (18.5-19.3%) and non-NWDAF baselines (3.6%). The results further indicate that the dual-agent design enables deeper and more uniformly distributed sleep opportunities while avoiding unstable transitions and load oscillations. These findings demonstrate that NWDAF-driven analytical abstraction combined with feasibility-aware, execution-centric learning provides a practical and scalable path toward green and load-stable 5G RAN operation.

키워드