Intelligent Control Systems for Industrial Processes Based on Machine Learning Methods

Abstract

This paper considers an approach to constructing intelligent control systems for industrial processes based on machine learning methods. The general architecture of such a system, the stages of training sample formation, and the integration of models into the loop of an automated process control system (APCS) are described. The role of the digital twin is highlighted as a safe simulation layer for verifying control recommendations, generating rare operating modes, and monitoring data drift. As an example, a calculated scenario for applying predictive models to short-term forecasting of process parameters and energy consumption is presented with indicative quality metrics. The expected implementation effects are discussed: improved control quality, reduced energy consumption and accident probability, as well as limitations and risks of the proposed approach.