| 杨乐* **,李萌* **,叶欣宇* **,孙恩昌* **,张延华* **.融合边缘计算与区块链的工业互联网资源优化配置研究[J].高技术通讯(中文),2020,30(12):1253~1263 |
| 融合边缘计算与区块链的工业互联网资源优化配置研究 |
|
| |
| DOI：10.3772/j.issn.1002-0470.2020.12.006 |
| 中文关键词: 工业互联网； 区块链； 深度强化学习（DRL）； 移动边缘计算（MEC）； 资源分配 |
| 英文关键词: industrial Internet, blockchain, deep reinforcement learning (DRL), mobile edge computing (MEC), resource allocation |
| 基金项目: |
| 作者 | 单位 | | 杨乐* ** | | | 李萌* ** | | | 叶欣宇* ** | | | 孙恩昌* ** | | | 张延华* ** | |
|
| 摘要点击次数: 2724 |
| 全文下载次数: 1664 |
| 中文摘要: |
| 随着通信网络技术的发展，工业互联网技术及其应用日益成熟。然而，工业互联网中对数据的安全性和隐私性需求为传统工业互联网架构带来前所未有的挑战。区块链技术作为极具发展前景的新技术之一，已被应用于工业互联网系统中。但是，当前基于区块链的工业互联网系统中仍存在一些亟待解决的问题，设备处理计算任务产生能耗较大，区块链中共识过程效率偏低，系统中存在严重的计算开销。针对上述问题，本文在工业互联网架构中引入移动边缘计算（MEC）技术，提升设备处理计算任务的能力和区块链节点的共识效率。同时，充分考虑工业互联网系统的设备能耗和计算开销。综上，本文提出一种融合边缘计算和区块链的工业互联网资源分配优化方法，以减少系统设备能耗和计算开销为目标，并将此优化问题构造为马尔可夫决策过程（MDP），系统中的卸载决策、区块尺寸和计算服务器均可动态调整和选择。根据优化场景的高动态、多维度特点，本文采取深度强化学习方法优化求解所提问题。通过仿真验证，相比于其他现有方法，本文所提方法可有效提升系统性能。 |
| 英文摘要: |
| Industrial Internet has emerged with the developments of various communication technologies. However, the requirement of data security and privacy in industrial Internet has brought unprecedented challenges to traditional architecture. As one of the promising technologies, blockchain has been applied into industrial Internet systems. Nevertheless, there are still some problems to be solved in blockchain-enabled industrial Internet systems, i.e. unbearable energy consumption for computation tasks, poor efficiency of consensus mechanism in blockchain, serious computation overheads of network systems. To handle the above issues and challenges, this paper integrates mobile edge computing (MEC) and blockchain into the industrial Internet systems to promote the computation capability of devices and improve the efficiency of consensus process. Meanwhile, the energy consumption and the computation overheads are jointly considered. Thus, a resource allocation optimization framework for blockchain-enabled industrial Internet systems is proposed to decrease energy consumption in devices and computation overheads of system. The optimization problem is formulated as a Markov decision process (MDP), and offloading decision, block size as well as computing server can be dynamically adjusted and selected. Accordingly, due to high-dynamic and large-dimensional characteristics of the system, the deep reinforcement learning is introduced to solve the formulated problem. Simulation results demonstrate that the proposed scheme can improve system performance significantly compared with other existing schemes. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
