Research of Surrogate Models Accuracy Based on Machine Learning for Adaptive Production Control

The accuracy of surrogate models based on machine learning methods designed for adaptive management of production processes as part of digital counterparts has been studied. A detailed simulation model of the production site has been developed in the AnyLogic environment, on the basis of which a data set has been formed for training and testing surrogate models based on the use of machine learning methods. The models are based on the following methods: random forest, gradient boosting and multilayer perceptron. The results of a comparative analysis of the accuracy and computational efficiency of the developed models in predicting the time characteristics of the production process are presented. It was found that the best accuracy is provided by the random forest model, for which the coefficient of determination in the test sample was up to 0.994, and the relative prediction error does not exceed 8%. It is shown that the use of surrogate models reduces the calculation time by more than 1800 times compared with the simulation model. Promising approaches to using ML surrogates as part of digital twins and multi-agent production systems forimplementing adaptive planning and optimizing production processes in real time are considered.