基于用户行为和位置感知的边际服务加载优化研究

发布时间：2018-03-12 23:02

本文选题：边缘计算　切入点：行为预测　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：边缘计算(Edge Computing,也称作移动边缘计算)是一种新兴技术,旨在为附近的移动用户提供云和IT服务。边缘计算平台通过在边际网提供计算和存储功能来降低网络延迟。它还使得应用开发者和内容提供者能够通过实时无线接入来提供上下文感知服务(诸如协同计算)。移动和物联网设备可以利用边缘计算服务为计算密集型应用(例如图像处理,移动游戏)实现计算迁移。边缘计算是位置信息敏感的云计算模式。因此,通过挖掘用户使用服务的行为模式并结合地理信息来预测用户的服务使用行为以及可能接入的Edge服务器,可令Edge服务器利用服务预加载来减少延迟,进而提升用户体验。然而,由于移动环境的上下文多变性导致用户行为的多变性,这使得移动用户行为模式的提取较之传统模式更为复杂。本文将移动用户所使用的应用作为服务,相关工作均以此展开。主要工作包括以下三点:1.根据收集到的真实用户服务使用记录构建边缘计算的场景,并分析了移动环境下用户的行为模式。利用word2vec词向量技术构建特征向量并利用LSTM建立基于上下文的序列预测模型。通过该模型来预测用户的服务使用行为,并结合相关概率模型选择相应的Edge服务器来预加载服务。2.根据用户使用服务的行为具有位置相关性的特点,结合地理POI信息构建Edge服务器地理位置特征,并利用BP神经网络挖掘Edge服务器位置与服务之间的映射关系,从而为服务预加载的优化提供更多选择。3.通过实验进行评估分析。实验结果证明了上述模型在一定程度上的有效性,为边缘计算优化问题提供了一种可参考的方案。
[Abstract]:Edge computing (also known as moving edge computing) is a new technology. It is designed to provide cloud and IT services to nearby mobile users. The edge computing platform reduces network latency by providing computing and storage capabilities on the marginal network. It also enables application developers and content providers to use real-time wireless. Access to provide context-aware services (such as collaborative computing). Mobile and Internet of things devices can utilize edge computing services for computation-intensive applications (e.g. image processing, Edge computing is a location-sensitive cloud computing model. By mining the behavior patterns of users using services and combining geographic information to predict service usage behavior and possible access to Edge servers, Edge servers can use service preloading to reduce latency and thereby enhance the user experience. The context variability of mobile environment leads to the variability of user behavior, which makes the extraction of mobile user behavior patterns more complex than the traditional mode. In this paper, the applications used by mobile users are considered as services. The main work includes the following three points: 1.Based on the collection of real user service usage records, we build edge computing scenarios. The behavior pattern of users in mobile environment is analyzed. The feature vector is constructed by using word2vec word vector technology and the context-based sequence prediction model is established by LSTM. The model is used to predict the service usage behavior of users. Combining with the correlation probability model, the author selects the corresponding Edge server to preload the service. 2. According to the characteristic that the user's behavior of using the service is location-dependent, combining with the geographic POI information, the author constructs the geographic location feature of the Edge server. BP neural network is used to mine the mapping relationship between Edge server location and service. The experimental results show that the above model is effective to some extent and provides a referential scheme for the optimization of edge computing.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.5;TP393.05

【相似文献】