A Modified Federated Singular Value Decomposition Method for Big Data and ML Optimization in IIoT Systems

Abstract

Continuous improvement of digital technologies allows for the introduction of new and improvement of existing areas of human life. Medicine, education, industry and many other areas have now, due to the pandemic situation, changed the usual way of functioning. The involvement of modern methods of data processing and management of large-scale systems allows for performing work tasks with less human intervention. The Industrial Internet of Things (IIoT) has become an important stage in the development of industrial systems. The organization of IIoT allows us to flexibly manage the production system, rationally use resources and provide users with a wide range of services. It is necessary to involve a large number of intelligent devices that process data from the environment and transmit it for further processing to control devices. For efficiently and quickly analyzing the information coming from different sensors we should use the methods of working with Big Data. Optimization of information that is not important for further analysis and decision-making can significantly reduce computational time while maintaining high accuracy. Machine learning methods simplify the process of Big Data analysis and create an industrial system based on the information obtained. Because IIoT systems are distributed and consist of many components, federated machine learning is used. The peculiarities of the use of distributed machine learning for Big Data analysis are investigated. Preliminary optimization of data arrays using the Singular Value Decomposition (SVD) method was proposed. The research results proved the proposed SVD method's effectiveness for Big Data pre-optimization in the IIoT systems. Also in the paper, we offered the modified method of SVD, which is adapted to work in the distributed IIoT systems. The research results showed high accuracy and reasonableness of calculations. To improve the efficiency of sparse data analysis in IIoT, it is also suggested to use the Funk-SVD algorithm.