In-network context inference in IoT sensory environment for efficient network resource utilization
In-network context inference in IoT sensory environment for efficient network resource utilization
复制标题
DOI：
10.1016/j.jnca.2019.01.013
复制
发表时间：
2019-03-15
影响因子：
8.7
通讯作者：
Saxena, Divya
中科院分区：
文献类型：
--
作者：
Verma, Rahul Kumar;Pattanaik, K. K.;Saxena, Divya
上下文感知是指应用程序从感测到的信息中理解情况并相应地以最少的人为干预提供服务的能力。感知上下文的物联网应用程序收集高级上下文（HLC），而不是来自传感器网络的原始传感器数据，以实时采取所需的操作。在低网络流量、延迟和能耗的情况下准确地推断上下文是一项具有挑战性的任务。现有的节点上数据处理机制只提供低级别的事件，不支持从多个传感器节点收集的多变量传感器数据的融合。另一方面，在网关处推断HLC需要将感测到的数据从网络边缘传输到网关（网络外），这导致高网络流量和能量消耗。在本文中，我们提出了一个朴素的计算轻量级的网络上下文推理机制，命名为InContextIoT，可以处理从不同的传感器收集的多变量传感器数据，用于有效地推断传感器网络内部的HLC。一个数据收集节点，命名为模式，选择的基础上，节点的剩余能量，接近中心，程度，和计算资源的可用性的HLC推理。然后，我们使用贝叶斯分类从收集的各种传感器节点的低级别事件中推断出模式下的HLC。Inode的选择是动态的，取决于应用程序查询的感兴趣区域（RoI）。实验表明，与现有的集中式最先进的方法相比，InConteactIoT通过将网络流量减少约两倍（2倍），将网络能耗降低到73%。
Context-awareness refers to the ability of an application to understand the situation from the sensed information and provides service(s) accordingly with the minimum human intervention. Context-aware IoT applications collect high-level contexts (HLCs) instead of the raw sensor data from the sensor networks to take the required actions in real-time. Inferring contexts accurately with low network traffic, latency, and energy consumption is a challenging task. Existing on-node data processing mechanisms provide only low-level events which do not support the fusion of multi-variate sensor data collected from multiple sensor nodes. On the other hand, inferring the HLCs at gateway, requires transmission of sensed data from the edge of the network to gateway (out-network) which incurs high network traffic and energy consumption. In this paper, we propose a naive computationally lightweight in-network context inference mechanism, named InContextIoT, that can process the multi-variate sensor data collected from different sensors for inferring the HLCs inside the sensor network efficiently. A data collection node, named Mode, is selected for the HLCs inference on the basis of node's residual energy, closeness centrality, degree, and availability of computation resources. Then, we use the Bayesian classification for inferring the HLCs at Mode from the collected low-level events of various sensor nodes. The selection of Inode is dynamic and depends on the region of interest (RoI) of application queries. Experiments show that InConteactIoT reduces network energy consumption to 73% by reducing the network traffic approximately two times (2x) in compared to existing centralized state-of-art approaches.