An Algorithm for Developing a Digital Twin Model of an Industrial Facility Segment

The article describes an algorithm for developing predictive models for a digital twin of a feedstock enrichment plant. The main results are summarized in this paper. The results obtained include: a generalized algorithm for the development of forecasting models, which can be used to create predictive models for digital twins using machine learning methods; the result of applying mathematical methods of preliminary selection of features for predictive models; the main results of the conducted experiments on the creation of predictive models based on linear regression without regularization, Lasso regression and random forest for regression; as well as predictive models obtained on a reduced feature space. The results of the experiments showed that the reducing number of features method taking into account the degree of their contribution to obtaining the predicted indicator, leads to a slight decrease in the quality of the model in accordance with the selected quality metrics; after screening out more than 70% of the initial features, it is possible to obtain interpreted models with a different composition of parameters and high accuracy.