A comparative analysis of methods for constructing mathematical models of object functioning using machine learning

The subject of the study is a technical object; its work is determined by many factors, its performance is characterized by some indicator. It is necessary to build a mathematical model that connects this indicator with the values of factors. As an example, the article examines the influence of various factors on the efficiency of burner devices (load, air consumption, methane and biogas, fuel and oxidizer compositions, and others). The efficiency (performance) of the burner device is assessed by the temperature of the flue gases. The problem is solved by machine learning methods, since classical regression analysis methods showed insufficient accuracy. The article explores the effectiveness of the following approaches: the support vector method, random foresting and decision tree boosting. The authors used a localized version 13.3 of the Statistica system for numerical calculations. All three machine learning approaches discussed in the paper have shown a significant increase in the model accuracy on the test sample. The method of boosting decision trees has shown the best results in this example. The recommended model construction technology that provides the necessary forecasting accuracy is first reduced to testing the classical regression analysis (if the resulting model provides the necessary accuracy, then it is preferable from the point of view of its interpretability). If the accuracy is insufficient, the three considered methods of machine learning are used. It this case, it is important to select the parameters of each of the methods, which, on the one hand, would provide the necessary accuracy, on the other hand, would not lead to model retraining. The resulting model can be used to assess the influence of various factors on the efficiency of the technical facility, as well as to predict its functioning quality (in particular in the considered example, to predict the temperature of flue gases).