基于深度强化学习的海洋移动边缘计算卸载方法  

Maritime mobile edge computing offloading method based on deep reinforcement learning

作　　者：苏新 孟蕾蕾 周一青[2,3] CELIMUGE Wu SU Xin;MENG Leilei;ZHOU Yiqing;CELIMUGE Wu(The College of IoT Engineering,Hohai University,Changzhou 231022,China;State Key Lab of Processors,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China;School of Computer Science and Technology,University of Chinese Academy of Sciences,Beijing 100190,China;Graduate School of Informatics and Engineering,The University of Electro-Communicaitons,Tokyo 182-8585,Japan)

机构地区：[1]河海大学物联网工程学院,江苏常州231022 [2]中国科学院计算技术研究所处理器芯片全国重点实验室,北京100190 [3]中国科学院大学计算机科学与技术学院,北京100190 [4]日本电气通信大学信息理工学院,东京182-8585

出　　处：《通信学报》2022年第10期133-145,共13页Journal on Communications

基　　金：国家重点研发计划基金资助项目(No.2021YFE0105500)。

摘　　要：海洋信息系统网络节点之间的强异构特性为海洋移动边缘计算任务卸载优化带来了复杂高维度的限制条件,同时复杂多样化的海事应用会导致海洋网络局部区域出现计算任务的超负荷处理。为实现海洋网络节点计算任务的最佳卸载与资源优化,满足网络低时延、高可靠的应用服务需求,提出基于多尺度异构特征属性的海洋网络节点分层归类方法和基于深度强化学习的海洋移动边缘计算卸载方法。仿真结果表明,所提方法较传统方法能够在海洋信息系统下有效地降低网络节点的计算任务卸载时延,并且能够在大规模任务流下保持海洋网络的稳健性。The strong heterogeneity among the network nodes of the maritime information system brings complex and high-dimensional constraints for optimizing task offloading of the maritime mobile edge computing. The complex and diverse maritime applications also lead to the overload processing of computing tasks in local areas of the maritime network. In order to optimize the task offloading and resource management of maritime network, as well as meet the maritime application service requirements of low-latency and high-reliability, a hierarchical classification method of maritime nodes based on multi-layers attributes and a novel offloading method for maritime mobile edge computing based on deep reinforcement learning were proposed. Compared with conventional methods, simulation results show that the proposed method can effectively reduce the computing task offloading delay of the marine information system, and maintain the robustness of the maritime network with large-scale task flows.

关 键 词：海洋信息系统 边缘计算 计算任务卸载 功率与计算资源分配 深度强化学习 

分 类 号：TN929.52[电子电信—通信与信息系统]