物联网资源管理框架及服务提供平台

发布时间：2018-06-27 20:06

本文选题：物联网 + 资源管理　；参考：《北京邮电大学》2014年博士论文

【摘要】：目前,物联网的应用模式还停留在Intranet of Things阶段,大多数物联网应用模式都是竖井式的,这种应用模式通常是：专用的设备和感知延伸网络用于专门的应用,底层的感知资源和感知数据被封锁在了一个个的应用中,得不到进一步的共享和重用。这些系统之间是相互隔离的,彼此的信息不能够互联互通,导致了信息孤岛的产生,造成了大量的传感资源和数据的浪费,也阻碍了进一步的跨系统的智能数据融合。此外,这种竖井式应用模式无法支持更加“普适性”和大规模的应用方案。本文主要关注的是支持从Intranet of Things过渡到Internet of Things所需要的中间件基础设施,以及构建这种基础设施存在的部分技术问题及其解决方案。本文提出了一个物联网服务提供框架。框架在保证资源提供者本身应用的实时性和可靠性前提下,为可定制的开放和共享资源提供了基础设施。框架中包含多级资源管理平台,作为资源管理平台的体系架构,而资源管理平台是框架的核心,作为整个基础设施的功能载体。从大的方面来讲,平台的目标是为从竖井式应用模式过渡到水平式模式提供基础中间件,进而以开放的方式将底层资源的能力提供出来,从而实现资源和数据的跨系统共享和重用。从功能上来说,平台提供了一个中间件来接入异构资源、形式化的描述资源和实体,并将它们的输出以"well-understood、machine-processible"的方式开放给不同的应用。针对物联网资源的异构性问题,资源管理平台主要解决协议的异构性和数据模型异构性两个方面。对于协议的异构性,平台底层借助于UDA (Unified Device Access)统一接入层来适配底层设备协议的异构性。UDA的核心功能是协议栈框架,协议栈框架基于OSGi技术,并采用Dependency-Inversion的协议栈设计模式。总体来说,协议栈框架有两大特点：支持协议的动态热插拔,即协议是即插即用的,可以动态的安装、修改和卸载；协议的自适配,即协议栈框架能够支持异构协议的自适配,能够自动选择协议栈或者动态组装协议栈来解析未知数据包,不需要手工逐一配置。此外,协议栈框架中不同的协议都以组件的形式维护在协议容器中,协议容器提供了维护、管理和配置协议的图形化管理工具,因此可以规范的管理已有协议。经过一系列的性能测试,实验结果表明UDA的性能可以胜任一些对实时性要求比较高的应用场景。对于数据模型的异构性,本文提出了基于语义本体的多层次、多维度信息模型。异构资源通过UDA接入进平台后,通过信息模型对资源和数据进行统一建模描述。多层次信息模型的理念首先从模型这个角度将上层应用与底层传感器资源解耦,上下层模型通过资源-实体绑定进行关联。资源和实体关联后,资源就可以观测或者控制实体的具体属性,原本无意义的观测数据也变成了具备具体应用意义的情景数据。一个应用场景中的实体模型可以关联不同资源提供者的资源模型,这意味着从模型的角度应用程序可以共享和重用传感器资源。同时,信息模型提供了模型的关联和映射机制,通过这种机制,系统之间的数据或者来自感知层的数据能够以上层应用可理解的方式提供给它们。此外,本文提出了多维度的思想,通过多个特征维度的资源模型来描述资源。根据多维度模型,资源匹配算法可以根据每个维度的特性准确的匹配资源。此外,我们实现了一个基于领域模板的建模工具,能够有效地创建、组织和维护这些模型并支持模板的复用。信息模型已在“西门子智能交通系统”等系统中得到了应用。平台接入大量的异构感知资源后,将它们能力以轻量级服务的方式开放出来。针对物联网服务的大规模性、动态性和情景的复杂性,本文提出了基于语义相似度的功能性服务推荐和基于协同过滤的QoS-based服务推荐算法,根据用户的需求,结合当前被监控实体的情景,为用户自动或者半自动的推荐合适的服务。算法的作用主要定位在两方面：协助用户从大量的可用服务集中选择合适的服务,降低用户的工作量；当资源或者被测实体情景发生改变,动态为被测实体匹配资源,找到与实体当前情景相符的感控服务,保证了服务的连续性。实验结果表明文中提出服务推荐算法是适合物联网环境特性的,它在较短的计算时问内尽量保证了推荐的准确率和召回率性能。
[Abstract]:At present, the application mode of the Internet of things is still in the Intranet of Things stage, most of the application modes of the Internet of things are vertical wells. This application mode is usually used for special applications and perceptual extension networks, and the underlying perceptual resources and perceptual data are locked in a number of applications. Sharing and reuse. These systems are isolated from each other, the information of each other can not be interconnected, resulting in the production of information islands, causing a large amount of waste of sensing resources and data, and also hindering further cross system intelligent data fusion. In addition, this type of vertical application model can not support more "universality" and larger. Scale applications. This article focuses on the middleware infrastructure needed to support the transition from Intranet of Things to Internet of Things, as well as some of the technical problems and solutions that exist in the construction of such infrastructure.
This paper presents a framework for the service of the Internet of things. The framework provides the infrastructure for customizable open and shared resources under the premise of the real-time and reliability of the resource provider itself. The framework includes a multilevel resource management platform as the architecture of the resource management platform, and the resource management platform is the core of the framework. As a function carrier for the entire infrastructure, the goal of the platform is to provide the basic middleware for the transition from the shaft type application mode to the horizontal mode, and then to provide the ability of the underlying resources in an open way, so as to realize the cross system sharing and reuse of resources and data. It provides a middleware to access heterogeneous resources, formally describe resources and entities, and open their output to different applications in a "well-understood, machine-processible" way.
In view of the heterogeneity of the Internet of things, the resource management platform mainly solves the heterogeneity of the protocol and the heterogeneity of the data model. For the heterogeneity of the protocol, the core function of the UDA (Unified Device Access) unified access layer to adapt the heterogeneous.UDA of the underlying device protocol is the protocol stack framework and the protocol stack. The framework is based on OSGi technology and adopts the protocol stack design pattern of Dependency-Inversion. In general, the framework of the protocol stack has two major features: dynamic hot plug in the protocol, that is, the protocol is plug and play, can be installed dynamically, modified and unloaded; the self adaptation of the protocol, that is, the protocol stack framework can support the self adaptation of heterogeneous protocols. Automatic selection protocol stack or dynamic assembly protocol stack to parse unknown data packets without manual configuration. In addition, different protocols in the protocol stack are maintained in the protocol container in the form of components, and protocol containers provide graphical management tools for maintenance, management, and configuration protocols, so it can be managed in a standardized way. After a series of performance tests, the experimental results show that the performance of UDA can be applied to some application scenarios with high real-time requirements.
For the heterogeneity of the data model, this paper presents a multi-level and multi-dimensional information model based on semantic ontology. After the heterogeneous resources are connected to the platform through UDA, the information model is used to describe the resources and data. The concept of multi level information model first from the angle of the model to solve the upper application and the underlying sensor resource solution. Coupled, upper and lower layers are associated by resource entity binding. After resources and entities are associated, the resources can observe or control the specific properties of the entity. The original meaningless observation data also becomes scenario data with specific application significance. An entity model in an application scene can relate to the resources of different resource providers. The model, which means that the application of the model can share and reuse the sensor resources. At the same time, the information model provides the association and mapping mechanism of the model, through which the data between the systems, or the data from the perceptual layer, can be provided to them in an understandable way. According to the multi-dimensional model, the resource matching algorithm can match the resources accurately according to the characteristics of each dimension. In addition, we implement a modeling tool based on the domain template, which can effectively create, organize and maintain these models and support the reuse of the template. The information model has been applied in the system of SIEMENS intelligent transportation system.
After the platform has access to a large number of heterogeneous perceived resources, the platform opens up their capabilities in a lightweight service. In view of the large-scale, dynamic and situational complexity of the Internet of things services, this paper proposes a functional service recommendation based on semantic similarity and a collaborative filtering based QoS-based service recommendation algorithm based on user requirements. The role of the algorithm is mainly in two aspects: assisting the user to select the appropriate service from a large number of available services, reducing the user's workload, and dynamically matching the measured entity when the resource or the measured entity situation changes. The results show that the service recommendation algorithm is suitable for the environment of the Internet of things, which ensures the accuracy and recall performance in the short calculation.
【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP391.44;TN929.5

【参考文献】