The Research Of Theory And Key Technology For Information Fusion In The Internet Of Things

After the development of computer and Internet, the Internet of Things (IoT) isconsidered as another great revolution in the information industry. The IoT hasbecome one of most popular research topics in the society due to its great potentialutilization in many different domains, such as retail business, pharmaceuticalsindustry, e-health, logistics, food industry, smart homes and intelligent transportation,and so on. With the development of intelligent sensing devices, manufacturingtechnology of sensors and communication technology, increasing amounts of sensingdevices (such as RFID, sensors and intelligent devices) are connected to the Internent.These devices generate and transmit data continuously. Therefore, there are hugevolumes of data crowding in the Internet. The IoT is realizing the transition fromstage of physical network, which consisted of intelligent sensing devices and theirnetworks for monitoring objective things, to stage of virtual resources network, whichincorporated data and information for reflecting objective things. Therefore, thesedata and information in the IoT is one of the key issues to achieve the goal ofubiquitous communication, comprehensive awareness and intelligent services.The IoT is a vital important part of the Future Internet. It is the infrastructures ofubiquitous communication among objective things. Further combined with netwroks,information, services and things will fully connect together. Hence, the substantivecharacteristic of IoT is ubiquitous communication. From the perspective of ubiquitouscommunication in the IoT, the IoT data has its own characteristics. Massive amountsof data: large volumes of data are continuously generated from a huge number ofsources along with increasing numbers of devices which are connected to the Internet. IoT data has high dimensional attributes for comprehensive awareness. Heterogeneousdata: In the IoT, varieties of devices are interconnected to perceive the objective world.Not only the data sensed to observe different objects may be heterogeneous, but alsothe data sensed by different kinds of sensors to monitor the same object may beheterogeneous. Real time: For monitoring objective things, the IoT data ischaracterized by time. Data distributed stored: These data is stored and maintained byorganizations that deploy the sensing networks. One of challenges of the IoT is toprocess information effectively.Information fusion is considered as a suit of strategy, which goal is to processinformation completely, accurately, timely and effectively. In the IoT environment,data fusion is also a framework that comprises theories, methods, and algorithms forinteroperating and integrating multisource heterogeneous data from sensormeasurements or other sources, combining and mining the measurement data frommultiple sensors and related information obtained from associated databases, andachieving improved accuracy and more specific inferences than that obtained by usingonly a single sensor. Research of information fusion focuses on architectures andfusion algorithms. The former is the guidance for costuming fusion strategy, and thelatter, keeping a certain data quality, is to simplify data and/or do data inference basedon some mathematic models.In this paper, some relevant research has been carried out in the following threeaspects:Firstly, focusing on multi-sources heterogeneous IoT data, architecture ofinformation fusion in the IoT is necessary to be proposed, which can illustrate overallguidance for the development of information fusion in the IoT. From perspective offunction, it can be seen as4phases: raw data acquisition, data abstraction, dataintegration and fusion, feature abstraction and inference. Workflow and functions ofevery phase is discussed in detail. Secondly, focusing on high dimensional attributes of IoT data，we propose anefficient feature fusion algorithm based on partitioning in general information systemto eliminate redundant data attributes. The basic concept involves the partitioning ofattribute dimensions, i.e., a big data set with higher attribute dimensions can betransformed into certain number of relatively smaller data subsets that can be easilyprocessed. Then, in general information system, every smaller data subset iscomputed. Based on computational results of the partitioning of dimensions, globaldata set are computed to obtain the result which is equal to the result by directlycomputing. It is a process of local calculations for the overall solution and it improvesthe computational efficiency of feature fusion in terms of high dimensional attributesof IoT data.Finally, focusing on massive amounts of IoT data, we propose an efficient fusionrules extraction algorithm based on the idea of positive approximation in incompletedecision information system. This algorithm is to extract simplified fusion rules i.e.relationships hided among objective things, for aiding decision making. In thisalgorithm, the idea of positive approximation is introduced into heuristic featureselection, which can remain results unaffected, diminish the computational universeof each granulation gradually. Therefore, this algorithm improves computationalefficiency of fusion rules extraction greatly.