2022
, Q4
Article
This study addresses the problem of improvement of the accuracy of the estimation of the quality of the end product when the used set of training data segments is small. It is extended with a physically grounded model (first principles model) of a mass-transfer technological process involved in the production of methyl tert-butyl ether (MTBE). It is shown that the proposed approach is superior to other methods for the construction of statistical models (soft sensors) to estimate the quality of the end products, since it makes it possible to take into account physically grounded relationships and characteristics of the technological process, which ultimately leads to an increase in the level of adequacy of the developed model. The feasibility of the use of a first principles model of the mass-transfer process in the algorithm for the development of statistical models to estimate the quality index of the end product are determined under conditions of parametric uncertainty in the Murphree efficiency of mass transfer and the parameters of the binary interaction between isobutylene dimers and components from the MTBE production system. The use of the area of the region of the intersection of output variable distributions between the values in the extended training and test samples is proposed as a criterion for the effectiveness of extension of the training sample.