ПОСТРОЕНИЕ МОДЕЛЕЙ ВИРТУАЛЬНЫХ АНАЛИЗАТОРОВ ЛЕГКИХ ФРАКЦИЙ ДЛЯ СИСТЕМЫ УПРАВЛЕНИЯ УСТАНОВКОЙ ГИДРОКРЕКИНГА НЕФТЯНЫХ ОСТАТКОВ
The problem of constructing virtual analyzers of laboratory indicators for light fractions of hydrocracking of oil residues is considered. In the course of performing this task, the main indicators of the state of the installation associated with the simulated laboratory values were identified. The task of constructing virtual analyzers was reduced to the classical task of "supervised learning" by shifting the set indicators to correctly selected time lags. An approach to estimating the values of time lags between independent variables and simulated values, based on cross-correlation analysis, is considered. The constructed models were tested on real data from the installation in Burgas (Bulgaria). The resulting virtual analyzers provide better estimates of model accuracy compared to typical analyzers built without taking into account linear regression lags.