物联网服务边缘适配与集成协同模式挖掘研究

With the rapid development and wide adoption of the Internet of Things (IoT), traditional device-centric IoT is moving into a new era where ubiquitous IoT resources are encapsulated and represented in terms of smart IoT services. In this setting, IoT resources at a certain network region are dynamically integrated through innovative IoT services for the realization of the value of interconnected IoT resources and the satisfaction of (near) real-time, intelligent and local user demands, and consequently, for the promotion of IoT intelligence at the edge of the network. In fact, the research and development of IoT services have become a recent hot topic for the inter-disciplinary exploration in IoT and service-oriented computing, and the manner of building novel IoT applications are changing. To address this challenge, IoT services network, which should be located at an edge node, is constructed leveraging the multi-dimensional and multi-granular IoT service specification, and considering the various types of relations and constraints derived from the structural and text description of IoT services. IoT services discovery and adaptability assessment mechanisms are proposed for facilitating the composition of IoT services at a certain edge node or contained at multiple neighboring edge nodes at the network edge. To promote the green IoT, sharable IoT services are identified for supporting concurrent IoT applications through the evaluation and optimization of the shareability of certain IoT services with respect to the requirement of IoT applications. Due to the multi-granularity of IoT services, the composition of IoT services should be represented as a layer-hierarchy. To promote the reuse and repurposing of IoT service (sub-)compositions, cross-composition and cross-layer-hierarchical fragments of IoT service compositions are extracted and evaluated for satisfying novel user demands, which may be partially similar to best practices. The user cases of smart toy computing are adopted for the evaluation and performance optimization of the techniques we have developed. Generally, this research aims to establish the scientific foundation and develop the key technical framework for assessing the adaptability of IoT services at the network edge and for extracting the most significant patterns of IoT service (sub-)compositions, and consequently, for laying necessary and solid theoretical and technical foundations about the design and development of novel service-centric IoT applications.

传统的以物理资源为中心的物联网正在向以泛在服务为中心的架构转变，通过把泛在资源虚拟化为物联网服务，在智能互联的网络边缘侧就近提供高效智能服务，边缘适配并动态集成协同服务以满足（准）实时智能化的本地应用需求，提升物联网边缘智能，已成为物联网和服务计算领域热点研究课题。在边缘计算情景下融合物联网服务的多维多粒度描述模型和服务之间存在的多元约束关系构建服务网络模型，提出面向物联网单和多应用的服务发现策略和边缘适配评估算法、以及服务与多应用之间可共享适应评估机制，实现多边缘网络环境下的服务动态集成协同和优化共享。依据层次性的物联网服务集成协同模式，挖掘跨协同模式、跨层次、适配可协作的协同模式片段，促进历史协同模式的重用和经改造后复用。采用智能玩具计算开展实例验证和效能优化。该研究将形成物联网服务边缘适配和集成协同模式挖掘的核心理论和关键技术框架，为新型物联网服务系统的设计与开发奠定理论和技术基础。

本项目探索物联网服务的边缘适配与集成协同优化，属于近年来服务计算和物联网交叉学科领域的热点研究问题。通过多应用集成优化促进资源的共享和最优配置，将传统的以物理资源为中心的物联网虚拟化为物联网服务，在靠近用户的网络边缘侧就近提供高效智能服务，以满足用户准实时智能化的本地应用需求。针对这些研究热点问题，本项目研究取得了如下进展：①提出了一种能量感知的服务组合策略，通过适配不同服务实现智能设备间的服务组合推荐；②提出了一种并发请求优化机制，通过整合可共享服务组件，优化并发任务结构，减少网络能耗；③提出了一种基于最小生成树算法的查询传输机制，优化边缘网络传输过程；④提出了一种基于边缘服务器协作的复杂任务在线调度机制，采用强化学习方法优化长期网络能耗和延迟；⑤在云边协同计算场景下，通过对不同服务类型进行划分，自适应地实现服务配置；⑥提出了一种以服务迁移为基础的服务重调度策略，通过服务按需迁移，实现计算资源优化调度。.作为物联网的一种特殊表现形式，本项目针对无线传感网进行了相关研究，并取得了如下进展：①提出了一种连续对象边界检测策略，实现工业无线传感网下的高精度边界区域检测；②提出了一种基于实时数据预测的高效事件确定方法，通过基于历史数据的建模来预测未来时间片数据，减少网络能量消耗延长生命周期。.此外，项目组还进行了图匹配的相关研究工作，取得了如下进展：①提出了基于快速归一化分割及乘法更新的图匹配方法，映射过程引入原函数的约束，所得到的解更贴合原目标函数要求；②提出了基于交替方向乘子法的图匹配方法，通过交替方向乘子法框架求得四个子问题的解，并协调得到原问题的解；③提出了基于邻接张量的子图匹配方法，将基于邻接张量的图匹配算法扩展至非同规模图匹配；④提出了一种基于概率的谱方法和乘法迭代更新的图匹配算法，在遥感图像匹配应用中，得到较高的匹配准确率。.该项目研究为物联网服务的边缘适配与集成协同奠定了较好的理论基础，所提出的核心算法和研发的原型系统为该项目学术成果走向实际应用积累了关键技术。作为上述研究的成果总结，申请人获得省部级一/二等奖2项，项目组成员在本领域知名国际期刊和会议上发表学术论文50篇，其中包括IEEE TSC、IEEE TII、IEEE IoT J等领域内知名期刊论文10篇、ICSOC、ICWS、ICKM等领域内知名国际会议论文11篇，培养博士研究生8名（已毕业2名）。