מושגי ליבה
A dynamic dormancy monitoring mechanism for Network Slicing-based IoT (NS-IoT) systems that leverages reinforcement learning to optimize energy consumption while maintaining required monitoring accuracy.
תקציר
The paper presents an innovative dynamic dormancy monitoring mechanism for Network Slicing-based IoT (NS-IoT) systems, which is built upon the NWDAF (Network Data Analysis Function) framework defined in 3GPP Release 17.
Key highlights:
- IoT devices are organized into groups, with each group monitored by a dedicated Monitor Application Function (MAF).
- A dormancy mechanism is introduced to place MAFs monitoring non-failing devices into a dormant state, reducing energy consumption.
- A reinforcement learning-based Proximal Policy Optimization (PPO) algorithm is developed to dynamically adjust the monitoring and dormancy durations for each MAF.
- The proposed approach maximizes energy conservation while maintaining the required monitoring accuracy, outperforming alternative strategies in terms of efficiency and stability.
The authors first introduce the NS-IoT system model and the dynamic dormancy monitoring mechanism. They then formulate the energy optimization problem as a Markov Decision Process (MDP) and solve it using the PPO algorithm. Simulation results demonstrate the effectiveness of the proposed approach in reducing energy consumption compared to full monitoring and other RL-based strategies.
סטטיסטיקה
The transmission power of IoTD m,n is pm,n.
The transmission rate vm,n for IoTD m,n is Bm log2(1 + pm,n/Nm).
The server processing power per unit time is edeal.
The server-cloud upload power per unit memory block is eup.
The abnormal power consumption of IoTD m,n during dormancy is el m,n.
ציטוטים
"Our goal is to obtain an optimal dormancy strategy to guide the sleeping and monitoring process for MAFm, so as to save total invalid energy consumption for long-term, which includes ineffective monitoring and abnormal device operation energy consumption."
"Deep Reinforcement Learning (DRL) algorithms have a significant advantage in solving such problems."