Year: 2025
Type: article
Abstract: The elemental data and mineral data obtained from elemental logging and X-ray diffraction were used to invert the mineral content through five methods. The oxide closure model method is based on the theoretical stratigraphic model, and the core condition of setting the sum of the mass percentages of clay minerals, carbonates, and the quartz-feldspar-mica mixture (QFM) to 100% is used for modeling and solving. The non-negative least squares method, truncated singu-lar value decomposition method, and linear programming method are based on the conversion relationship model between elements and minerals proposed by Herron. Through the analysis of the lithological characteristics of the block, the element content inversion mineral content model is established based on the matrix solution method. The XGBoost machine learning algorithm automatically learns and captures the complex relationships between samples from the data samples to solve the problem. Taking well x as an example, the mineral content was calculated through the five inversion methods. The com-parison of the cal-culation results with the whole rock analysis results shows that the variation trends of each method with the depth of the stratum are basi-cally consistent with the whole rock analysis results. Among the four mineral inversion meth-ods, the processing result of the XGBoost machine learning algo-rithm is the best, with a mean square error of 0.91; the processing effects of the non-negative least squares method and the truncated singular value decomposi-tion method are second, with mean square errors of 0.88 and 0.81, respectively; the mean square error of the linear programming method is relatively low, only 0.79.
DOI: 10.63313/ajet.9010
Author: Ruonan Zhang
Institution: Xi'an Shiyou University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4415141069