Research And Design Of Efficient Reasoning Over IoT Knowledge Sub-System

With the rapid development of the Internet of Things technology,its application fields have covered all aspects of people's lives,and physical devices generate a large amount of data,which forms event streams.The original complex event processing system is only to analyze the relationship between events.But the relationships between the attributes of physical devices hide behind the events.If these relationships are used,more rich and valuable high-level information and knowledge can be obtained.Therefore,the ontology is used to represent the resources,attributes and relationships of the Internet of Things,and the knowledge sub-system is obtained.In this way,not only the resource model of the Internet of Things,but also a large number of event streams need to be processed,and the reasoning efficiency is seriously affected.This thesis improves the efficiency of logical reasoning through parallelization methods.Its main work and contributions include the following aspects.(1)The segmentation of the knowledge set.The efficiency of parallelization is improved by segmenting the knowledge set,and analyze the satisfiability after segmentation.The calculation results of the satisfiability before and after the segmentation are consistent.Analyze the effects of different segmentation methods on efficiency.(2)Solving the space.By cutting the solution space,each computing unit only needs to search in a smaller space.Cut the solution space by using three different methods,statistics,random,frequency.Analyze the parallel efficiency.Select the appropriate cutting method.And the dynamic secondary cutting solution space is carried out in the calculation process to balance the calculation pressure of each calculation unit.(3)Inference testing based on minimum information exchange.Each calculation unit does not exchange all the solution information,but only selects the shared proposition variable,and selects the appropriate range to exchange the solution information to reduce the information exchange between each calculation unit as much as possible.And implement parallelization reasoning based on this design.FiInally,by deploying computing nodes on cloud platform,a large-scale parallel test under 400 computing nodes is implemented to verify the availability and efficiency of parallel reasoning.