Energy-Aware Design Policy for Network Slicing Using Deep Reinforcement Learning
Ranyin Wang, Vasilis Friderikos, A.H. Aghvami · IEEE Transactions on Services Computing · 2024
Network slicing technology promises to be a critical enabler of fifth-generation (5G) and sixth-generation (6G) wireless networks, allowing the infrastructure to be divided into several virtual slices. Mobile network operators are focused on developing novel solutions to implement various slices to address diverse use cases and guarantee key performance indicators (KPIs) such as latency, resource utilization, and energy efficiency. Energy efficiency (EE) KPIs are particularly important during slice deployments to ensure a reduced carbon footprint. However, deploying slices with high energy efficiency presents significant challenges. This paper proposes an energy-aware design policy for deploying slices by optimizing energy consumption and deployment capacity. A deep reinforcement learning (DRL) approach is employed, utilizing an actor-critic architecture to train a learning network modeled with a pointer network structure and an attention mechanism. A search strategy is also proposed to refine learning parameters and determine the final design policy. Compared to two existing approaches, the proposed algorithms demonstrate improved performance in terms of energy efficiency and cumulative acceptance ratio. Specifically, the EE KPI achieved by the proposed approach is enhanced to 69.7%.