METHOD OF MODEL BUILDING FOR ESTIMATION OF QUALITY PARAMETERS OF FRACTIONATION COLUMN PRODUCTS UNDER CONDITIONS OF SMALL VOLUME OF ANALYTICAL CONTROL DATA

The problem of improving the accuracy of models for estimating the low-temperature properties, flammability and anti-wear properties of the target products of the fractionation column under conditions of a small amount of analytical control data is considered. For the solution of the considered problem the method of model building is proposed, which includes the algorithm of expansion of a small training sample on the data of fractional composition, differing in the way of selection of additional data, taking into account the sparsity indicator, which allowed to include the missing amount of data in the training sample, and as a result to ensure the improvement of the model quality. The use of the proposed method improved the accuracy of the models by 18% on average compared to the known methods and by 6% on average compared to the method based on the expansion of the training sample without taking into account the sparsity index. The results are presented on examples of model building of quality indicators of filterability limit temperature, flash point, kinematic toughness at 40°C and cetane number of middle distillate (diesel fuel fraction) and flash point of kerosene fraction of industrial fractionation column of hydrocracking process unit.