A high-precision, complex, three-dimensional (3D) geological model can directly express the attributes of stratum thickness, geological structure, lithology and spatial form, which can provide a reliable basis for the development and utilization of underground space and planning decisions. However, it is difficult to perform accurate modelling due to the lack of basic data. As such, this paper proposes coupling a machine learning algorithm (K-nearest neighbour (KNN)) with the kriging algorithm to construct the topological relationship between the Delaunay triangle and the Thiessen polygon in order to perform the simulation and prediction of virtual drilling. Based on KNN, support vector machine (SVM) and neural network algorithms as well as the virtual borehole encryption data, data standardization processing and analysis are carried out. Through model verification, algorithm optimization is realized, and the optimal modelling method is explored. The results show that the fine KNN algorithm improved by Bayesian optimization can effectively improve the modelling accuracy through 0.1-m encryption, standardization processing and 5-fold cross-validation. Stratum modelling combined with the fine KNN and kriging algorithms can obtain a more accurate modelling without adding virtual boreholes. The improved levels of upper and lower hybrid modelling with an appropriate number of profile boreholes can also effectively optimize model accuracy. Both modelling accuracy and efficiency can be significantly improved by using Delaunay triangles and Thiessen polygons with virtual boreholes. Stratum modelling can effectively express the geological pinch-out in areas with adequate degrees of stratification, and hybrid modelling performs well in irregular geological bodies such as karsts and lenses.
Read full abstract