Intelligent Decision-Based Edge Server Sleep for Green Computing in MEC-Enabled IoV Networks

With the advent of the 6G era, addressing the increasing demand for mobile computing has become crucial. In the next-generation Internet of Vehicles (IoVs), widespread deployment of edge servers ensures low delay and high reliability for vehicular applications. However, densely deployed edge servers are designed to accommodate peak traffic. This results in unnecessary resource wastage, high maintenance costs, and energy consumption during off-peak hours. Enabling the sleep function of edge servers and dynamically controlling their operating mode is crucial for reducing energy consumption and enabling green IoVs. In this paper, we study the dynamic sleep problem of edge servers in the IoV network and propose traffic-aware intelligent sleep decision-making algorithms based on Deep Reinforcement Learning (DRL). We propose a Centralized DRL-based Dynamic Sleep (CDDS) algorithm, which leverages the deep deterministic policy gradient algorithm to learn the optimal decision policy through environmental interaction. To enhance the stability of agent learning and mitigate the impact of environmental changes, we propose a baseline transformation strategy based on the greedy algorithm. Additionally, to overcome the limitations of CDDS, we combine federated learning with DRL and propose a Federated DRL-based Dynamic Sleep (FDDS) algorithm. This approach speeds up model training and improves the model's generalization ability. Furthermore, we conduct extensive experimental verification using real-world datasets. The experimental results demonstrate that both CDDS and FDDS successfully learn the optimal sleep control policy, leading to system cost reduction of 11.06% to 45.36% and 8.71% to 43.77%, respectively, compared to the baselines.