Modeling of high-dimensional systems based on graph condensation

To address the challenges of planning and managing organizational, technical, and economic structures, a new methodology for processing large factor models is proposed, based on graph condensation. This methodology involves restoring the dependencies between factors using machine learning techniques and artificial neural networks. For creating a condensed graph, approaches based on both content analysis and methods for identifying strong components are explored. Subsequently, the vertices of the graph are clustered based on defining characteristics that describe them. The factor values for the vertices of the condensed graph are determined using a system of linear methods that describe mutual influence. Unknown schemes dependent on parameters within a single cluster are controlled using machine learning methods. These schemes are processed independently, allowing their implementation in a hybrid computing system in parallel mode. The mathematical methods used in this work are not original per se, but their novelty lies in their integration within a unified model for processing high-dimensional graphs. The proposed methodology is not tied to a specific subject area and is aimed at modeling simple organizational and technical systems when detailed behavior models of individual elements are absent, but there are small expert assessments of the degree of influence of industrial factors on one another, as well as some datasets on past interaction precedents among elements. The methodology for application to restore the factors of economic activity of individual economic entities is considered at the level of specified growth rates of industries, scientific and production enterprises, organizations, workshops, and laboratories.