ACCURACY AND RELIABILITY ASSESSMENT OF ENGINEERING GEOLOGICAL MODELS BASED ON MACHINE LEARNING

Assessing reliability of engineering geological models still requires further studies.. Much more attention is paid to the methodology for assessing the reliability and quality of models in oil and gas geology. A comparison of the methods used in this field with the methodology for assessing the quality of machine learning models showed the similarity of principles and approaches. Therefore, the algorithms for engineering geological modeling can be justified and calibrated using tools for evaluating the quality of machine learning models. The article systematizes and analyzes the metrics used in solving problems and classification, and describes the methodology of using cross-validation techniques to assess the quality of algorithms. The practical experience of constructing a computer stratigraphic-genetic model using various algorithmic approaches is described: on the basis of triangulation constructions, using ordinary kriging with automated variogram composition and using the custom heuristic algorithm that considers the history of sedimentation and technogenic transformation of the territory. It is shown that the problem of three-dimensional geological modeling can be considered both as a classification and regression problem. The error index of the stratigraphic-genetic model is proposed based on the calculation of the average absolute errors in determining the spatial position of geological layers. The proposed approaches are applicable to testing methodologies of engineering geological modeling in a broad sense, the verification of predictive models of any kind being the most difficult issue. It is emphasized that the development and filling of databases of various engineering and geological data should be intensified, including field and laboratory data, the results of their processing, forecast estimates and conclusions based on them, monitoring measurements of various kinds, remote sensing data, etc. The possibilities of processing and analyzing big data in engineering geology will allow us to move from subjective expert estimations to the application of modern approaches to modeling complexly formalized objects and phenomena using the capabilities of machine learning and artificial intelligence.