Application of machine learning algorithms in predicting pyrolytic analysis result

Abstract

Introduction. Geochemical studies of organic matter in oil source rocks play an important role in assessing oil and gas accumulation in any territory. These studies play a particularly important role in forecasting unconventional resources and oil and gas reserves (so-called shale hydrocarbons). It is recommended to carry out pyrolytic studies by the Rock-Eval method for rocks saturated with organic matter on samples before and after their extraction with chloroform. However, extraction is a laborious and time-consuming process, and the load on laboratory equipment and the time required for analysis is doubled.Aim. To get a working model for predicting pyrolytic parameters of extracted samples, without carrying out extraction analysis.Materials and methods. In this paper, machine learning regression algorithms are applied for predicting one of the pyrolysis parameters of extracted samples based on the pyrolytic analysis results of the extracted and non-extracted samples. To develop the prediction model, 5 different machine learning regression algorithms were applied and compared, including multiple linear regression, polynomial regression, support vector regression, decision tree, and random forest.Results. The prediction result showcases that the relationship between the parameters before and after extraction is complex and non-linear. Some methods have shown their incompatibility with the assigned tasks, others have shown good and satisfactory results. Those algorithms can be applied to predict all geochemical parameters of extracted samples.Conclusions. The best machine learning algorithm for this task is the Random forest.