INTELLECTUAL METHODS OF MODELING AND MANAGEMENT THE REFINING PROCESS BASED ON VIRTUAL ANALYZERS

Abstract

The article examines an approach based on the construction of intelligent virtual analyzers, which calculate the values of difficult-to-measure quality indicators based on the current readings of technological parameter sensors. A methodology for forming such models under conditions of noise, incomplete, and temporarily dissynchronized data is proposed. A probabilistic maximum algorithm is used to clarify time delays, partial least squares regression for working with multicollinear inputs, nonlinear transformations, analysis methods, and training sampling formation. It has been shown that the implementation of the developed virtual analyzer allows for a decrease in the average error of quality forecasting by tens of percent, stabilizes the refining regime, and ensures a reduction in energy consumption and losses of the target product.