MATRIX FACTORIZATION OF BIG DATA IN THE INDUSTRIAL SYSTEMS

Abstract

The creation of new technologies and their implementation in various fields necessitated Big Data processing and storage. In industrial systems, modernization means the use of a large number of smart devices that perform specialized functions. Data from such devices are used to control the system and automate production processes. A change in the parameters of individual components of the manufacturing system may indicate the need to adjust the global management strategy. The intelligent industrial systems main characteristics were defined in the paper. The Industrial Internet of Things concept and the relevance of the modernization problem for manufacturing were analyzed. The problems of processing Big Data in Industrial Internet of Things systems were examined in the paper. The use of recommendation systems for quickly finding relationships between users and production services was considered. The advantages of Big Data analysis by recommendation systems, which have a favourable effect on industrial enterprise efficiency were given. The use of SVD and FunkSVD matrix factorization algorithms for processing sparse data matrices was analyzed. The possibility of optimizing arrays of information, choosing the most important, and rejecting redundancy with the help of the above algorithms was determined. The proposed algorithms were simulated. The advantages of FunkSVD for working with sparse data were assigned. It was found that the FunkSVD algorithm processes the data in a shorter time than SVD, but this does not affect the accuracy of the result. The SVD is also more difficult to implement and it requires more computing resources was established. It has been shown that FunkSVD uses a lot of data to determine the relationships between it and make recommendations about the products most likely to be of interest to users. To increase the efficiency of processing large sets of information the FunkSVD algorithm was improved in such a way that it uses fewer data to generate recommendations. Based on the results of the research, the modified method works faster than the non-modified one but retains high calculation accuracy, which is important for work in recommender systems. The possibility of providing recommendations to users of industrial systems in a shorter period, thus improving their relevance, was revealed. It was proposed to continue research for finding the optimal parameters of the FunkSVD algorithm for Big Data processing.