计算机科学
切片
移动边缘计算
边缘计算
任务(项目管理)
移动计算
计算机网络
GSM演进的增强数据速率
分布式计算
计算卸载
移动电话技术
服务器
移动无线电
电信
万维网
管理
经济
作者
Xulong Li,Wenzhe Mao,Xinyi Xu,Yaxi Liu,Huifeng Zhang,Wei Huangfu,Keping Long
标识
DOI:10.1109/tccn.2025.3565942
摘要
Unmanned aerial vehicle (UAV)-enabled mobile edge computing (MEC) has received wide attention for its ability to significantly reduce latency with flexible mobile aerial nodes. However, the existing works merely consider a fixed periodic network slicing and usually neglect the characterization of computing tasks in practical applications. Motivated by this, we envision a novel scenario where the dynamics and heterogeneity of users and tasks are considered in UAV-enabled MEC networks. We propose a novel self-adjusting two-timescale network slicing scheme based on the internal mechanism in deep reinforcement learning (DRL) to automatically determine whether and when network slicing needs to be reconstructed. Such a novel scheme balances reconstruction cost and system performance compared with traditional fixed schemes. We build an optimization to minimize total system cost for joint UAV trajectory planning and resource allocation at both network and user slice levels. To tackle this problem, we design a novel self-adjusting two-timescale proximal policy optimization (PPO) algorithm that utilizes the property of the critic network, referred to as SaTPPO. Experiment results manifest the effectiveness of the proposed SaTPPO, and it has a lower system cost than the traditional fixed schemes. Moreover, the proposed SaTPPO outperforms the state-of-the-art baselines concerning effectiveness and efficiency.
科研通智能强力驱动
Strongly Powered by AbleSci AI