| In recent years,with the continuous expansion of the application field of the Internet of Things and the explosive growth of the deployed Internet of Things devices,massive multi-source heterogeneous Internet of Things data analysis methods have become the current research hotspot.Tensor decomposition has become one of the mainstream tools for data analysis in the Internet of Things due to its characteristics of mathematical interpretation of the correlation between multidimensional heterogeneous data.In this paper,the analysis method of streaming tensor is studied for the application scenario of continuous output streaming tensor source devices in the Internet of Things.Firstly,this paper proposes a unified description model and method for heterogeneous data based on tensor extension operator.Further,on the basis of in-depth study of CANDECAMP/PARAFAC(CP)decomposition and Tucker decomposition algorithms as well as traditional streaming tensor analysis methods,this paper proposes windowbased dynamic streaming tensor analysis method based on CP decomposition(CP-WTA)to analyze streaming data in the single-channel Internet of Things.Multi-streaming tensor fusion analysis(MSTFA)method and time-space joint multi-streaming tensor analysis(MSTA)method are proposed to meet the requirements of multi-channel Internet of Things streaming tensor analysis scenario.The main research contents and innovative achievements of this paper are as follows:1.The data of the Internet of Things has the characteristics of mass,polymorphism and relevance.According to the above characteristics,this paper proposes a data description model based on tensor extension operator,which can realize the unified description and fusion of structured,semi-structured and unstructured data of the Internet of Things.2.CP-WTA is proposed to meet the demand of local analysis of streaming tensor in a certain period of time in the streaming tensor scene of single-channel Internet of Things which includes independent-window-based streaming tensor analysis(IWSTA)algorithm and sliding window-based streaming tensor analysis(SWSTA)algorithm.The experimental results show that the execution speed of SWSTA method is 3-8 times higher than that of IWSTA method,and it has the ability of fast tracking,which can realize the abnormal trend detection of multi-dimensional data association relationship.3.In recent years,the research work of streaming tensor analysis focuses on the scene of single streaming data,but fails to take into account the typical needs of multiple data streams at the same time and in different spaces for fusion analysis in the Internet of Things.Therefore,this paper proposes time-space joint multiple-streaming tensor analysis(MSTA)method.The experimental results show that MSTA can realize the fusion of multipath streaming tensor and adjust the projection matrix smoothly when the multipath streaming tensors arrive,MSTA also can incrementally extract the core data of the multipath streaming tensor,thus realizing the analysis of the correlation mode of multidimensional heterogeneous data in the multipath Internet of Things data. |
