MEC环境中面向5G网络切片的计算卸载方法

Computation Offloading Towards 5G Network Slicing in Mobile Edge Computing

5G网络切片与计算卸载技术的出现,有望支持移动边缘计算(Mobile Edge Computing,MEC)系统在降低服务延迟的同时提高资源利用率,进而更好地满足不同用户的需求.然而,由于MEC系统状态的动态性与用户需求的多变性,如何有效结合网络切片与计算卸载技术仍面临着巨大的挑战.现有解决方案通常依赖于静态网络资源划分或系统先验知识,无法适应动态多变的MEC环境,造成了过度的服务延时与不合理的资源供给.为解决上述重要挑战,本文提出了一种MEC环境中面向5G网络切片的计算卸载(Computation Offloading towards Network Slicing,CONS)方法.首先,基于对历史用户请求的分析,设计了一种门控循环神经网络对未来时隙的用户请求数量进行精确预测,结合用户资源需求对网络切片进行动态调整.接着,基于网络切片资源划分的结果,设计了一种双延迟深度强化学习对计算卸载与资源分配进行决策,通过解决Q值过高估计和高方差问题,进而有效逼近动态MEC环境下的最优策略.基于真实用户通信流量数据集,大量仿真实验验证了所提的CONS方法的可行性和有效性.与其他5种基准方法相比,CONS方法能够有效地提高服务提供商的收益,且在不同场景下均展现出了更加优越的性能.

The emerging technologies of 5G network slicing and computation offloading will support Mobile Edge Computing(MEC)systems to reduce service latency and improve resource utilization,and thus the requirements of different users can be better met.However,due to dynamic system states and variable user requirements,it is challenging to effectively integrate network slicing and computation offloading technologies.Existing solutions commonly rely on static network resource partitioning or system priori knowledge,which cannot well adapt to such dynamic and variable MEC environments,resulting in excessive service delay and unreasonable resource provisioning.To address the above important challenges,we propose a Computation Offloading towards Network Slicing(CONS)method in MEC environments.First,based on the analysis of historical user requests,a gated recurrent neural network is designed to accurately predict the number of user requests for future time slots,which can be used to dynamically adjust the network slicing referring to user requirements.Next,based on the partitioning results of network slicing resources,a twin delayed deep reinforcement learning is designed to make decisions of computation offloading and resource allocation,which can approximate the optimal policy in dynamic MEC environments via solving the problems of Q-value overestimation and high variance.Based on the real-world datasets of user communication traffic,extensive simulation experiments are conducted to validate the feasibility and effectiveness of the proposed CONS method.Compared with five state-of-the-art methods,the CONS method improves the profit of service providers and exhibits superior performance in various scenarios.