项目来源

台湾省政府科研基金(GRB)

项目主持人

张耀中

项目受资助机构

台湾省台东大学资讯工程学系

项目编号

MOST107-2221-E143-001-MY3

财政年度

2020,2019

立项时间

未公开

项目级别

省级

研究期限

未知 / 未知

受资助金额

1366.00千元台币

学科

资讯科学——软体

学科代码

未公开

基金类别

应用研究/学术补助

行动边缘计算 ; 集中式云端服务 ; 边缘云端存取服务 ; 深度学习平台 ; 行动边缘雾运算闸道器 ; 流量卸载功能 ; Mobile Edge Computing(MEC) ; Central Cloud Service ; Edge Cloud Access Service ; Deep Learning Service Platform ; MEC Fog Gateway ; Traffic Offload Function(TOF)

参与者

赖盈勳

参与机构

未公开

项目标书摘要：随着行动互联网和物联网(IoT)应用的迅速发展，传统的集中式云计算面临着高延迟、低频谱效率与非自适应机器类型通讯的严峻挑战。为了解决这些挑战，新发展之网路通讯技术正由集中式云端计算转移至网路边缘计算之趋势。本计划首先探讨当前三种行动边缘计算技术，目标以发展减少延迟、提高频谱效率之B5G行动边缘雾运算网路架构，并支援大规模机器类型通信架构。藉由所规划之B5G行动边缘雾运算网路架构，包含Central Cloud Layer与Edge Cloud Access Layer,实作B5G行动边缘雾运算闸道器(MEC Fog Gateway),以强化单一装置及区域的运算效能、储存资源与资料处理能力，缓和传统云端运算架构之边缘计算所遭遇之资源不足及改善服务等待时间过长等问题。Edge Cloud Access Layer之MEC Fog Gateway运用Central Cloud Layer之巨量资料深度学习平台所建立之机器学习模型进行深度学习，由不同种类流量蒐集特徵资讯形成大数据资料，再将特徵资讯分成两部分，作为独立训练资料集与测试资料集。Central Cloud Layer之巨量资料深度学习平台透过Edge Cloud Access Layer之MEC Fog Gateway输入训练资料集，学习模型可自动调整类神经网路节点间权重值以及偏移量，使得计算训练阶段之误差能达到最小值。MEC Fog Gateway再以测试资料集作为输入，透过与已训练之神经网路节点权重值与偏移量进行计算，确认模型预测流量特徵，藉此验证学习模型之预测准确率。此外，本计划藉由优化B5G行动边缘雾运算智慧学习网路环境，提出MEC Fog Gateway协同管理服务与动态服务流量分流机制，以无线网路讯息模组(RNIS)与流量卸载功能模组(TOF)为基础，透过 MEC Fog Gateway之深度学习特性，即时分析MEC Fog Gateway执行效能以配置相应之分流策略，使MEC Fog Gateway可依据分流策略达到最佳运作效益，并有效降低整体网路延迟率及减少整体服务之封包遗失率，提升B5G行动边缘雾运算网路服务品质。

Application Abstract: With the rapid development of mobile Internet and Internet of Things(IoT)applications,the traditional centralized cloud computing faces the severe challenge of high latency,low spectral efficiency,and non-adaptive machine type communication.The emerging network communication technology is shifting from the centralized cloud computing to the network edge computing for addressing these challenges.The project first explores the three current operational Edge Computing(EC)technologies.With the goal of reduced latency and spectral efficiency and support for large-scale machine type communications architectures,this project develops the B5G Mobile Edge Fog Network architecture.The proposed B5G Mobile Edge Fog Network architecture includes the Central Cloud Layer and the Edge Cloud Access Layer.The B5G MEC Fog Gateway is implemented to enhance the computational efficiency of a single device and area,storage resources and data processing capabilities to ease the edge of the traditional cloud computing infrastructure suffered the problem of insufficient resources.The MEC Fog Gateway uses the machine learning model established by the deep learning platform to collect characteristic information from different types of traffic.Then the traffic of Internet divides into two parts as independent training data set and test data set.The deep learning platform input training dataset through the Edge Cloud Access Layer MEC Fog Gateway.The learning model automatically adjusts weights and offset between neural network nodes.The MEC Fog Gateway then uses the test dataset as input to verify that the prediction model by calculating weights and offsets from the trained neural network nodes.Finally,this project presents the Collaborative Management Service(CMS)based on MEC Fog Gateway by optimizing B5G Mobile Edge Fog Computing Intelligence Learning Network environment.The proposed Dynamic Service Traffic Diversion mechanism(DSTD)contains the Wireless Network Message Module(RNIS)and Traffic Offload Function Module(TOF).The MEC Fog Gateway performs the analysis of execution performance instantaneously to configure the corresponding traffic diversion strategy through the Deep Learning feature.Eventually,the MEC Fog Gateway achieves the optimal operational benefits and reduces the overall network latency and the total packet loss rate.Ultimately,this project will provide the service quality of the B5G Mobile Network Edge Fog Computing environment.

项目受资助省

台湾省