Synthesis of control systems with a predictive model by a group of specialists of different profiles

Abstract

The study focuses on systematizing the solution to the problem of choosing the structure of the predictive model in management. In the introductory part of the work, approaches to modeling based on requirements and data are considered. Common ways to reduce the influence of the model forecast bias on the quality of control are considered. It is shown that rapid changes in the state of the process require attention to the problems of the model itself. The problems are caused by the properties of retrospective data on the technological process under conditions of disturbances. The paper considers the problem of the human factor in the structural identification of models due to the involvement of specialists from different areas in the process of synthesis of the control system and the model itself. At the same time, it is assumed that the specialists can work asynchronously during different time intervals. With this in mind, we determined the stages of the activities of the specialists in the synthesis of predictive models in the context of the concepts of digital twins and holonic control. The proposed activity stages demonstrate the antagonism between data-based and requirements-based modeling approaches and define ways to reconcile model properties with modeling goals. The conclusions contain two examples demonstrating the use of a data-driven approach in situations where the use of a requirements-based approach is difficult. In the examples, the process of improving the model and coordinating it with the management tasks is considered. The first example is devoted to the problem of quality control of iron ore sinter. The second example deals with the temperature control of steel strip in hot dip galvanizing.