Research On Edge Data Management Architecture And Algorithms In Industrial Internet

With the development of emerging information technologies such as the Internet of Things,5G,artificial intelligence and big data,data-intensive industrial applications usually require large-scale time series data to support real-time processing and analysis.The traditional cloud computing model cannot be applied to all industrial application scenarios due to its high response delay and network bandwidth requirements.The edge computing model can effectively solve the above problems.However,the limited computing and storage resources of the edge computing nodes cannot deal with the massive computing and storage requirements of industrial applications alone.Therefore,the storage management of large-scale industrial time series data has become a critical problem that needs to be solved urgently.In order to solve this problem,this paper uses machine learning methods and data reduction techniques to propose an edge data management architecture and corresponding algorithms that support edge-cloud collaboration.The detailed research contents are summarized as follows:According to the characteristics of industrial application scenarios,on the basis of the open source edge computing framework Edge X,an edge data management architecture supporting edge-cloud collaboration is proposed.This article first studies the basic structure and implementation mechanism of the open source edge computing framework Edge X.Secondly,it designs the edge data management architecture and its main components that support edge-cloud collaboration.Finally,it describes the operating mechanism and deployment methods of the core components in detail.Aiming at the problem of different access requirements for time series data in different industrial applications,a data partition algorithm(DPM)based on application access patterns and time series data characteristics is proposed.Firstly,the correlation model between the frequency of data access by the edge or cloud industrial application and the characteristics of the accessed data is established.Secondly,based on the model,a data partition algorithm based on prediction is proposed to store the data separately according to the requirements of edge and cloud applications.Finally,a dynamic adaptive selection mechanism of prediction method is proposed to improve the accuracy of DPM.Aiming at the problem of limited storage resources at edge nodes,a data reduction algorithm(CDRM)based on prediction combining change point detection and time series split is further proposed.First,a basic data reduction algorithm(BDRM)based on exponential smoothing prediction method is proposed to reduce the edge storage overhead by retaining only part of the data instead of all the data.Secondly,in order to solve the problem that BDRM has a poor effect on the sequence reduction of non-stationary time series data,a comprehensive data reduction algorithm(CDRM)is proposed by introducing the change point detection and time series split technology to improve the reduction effect of non-stationary time series data.Finally,this paper uses real industrial data to evaluate and analyze the designed edge data management architecture and the performance of the proposed algorithms.The experimental results verify the effectiveness of the proposed architecture and related algorithms.