Adaptive segmentation algorithm for 3D geological modeling of long-distance lines

In 3D geological modeling of long-distance lines, building a model for the entire area is often a lengthy process. Moreover, the reliability of the model is sometimes uncertain. These problems arise because the line covers a large region, involves complex strata, shows strong stratigraphic variations, and passes through areas with large elevation changes. A practical solution is to divide the line into smaller segments for modeling. This study introduces an adaptive segmentation algorithm that integrates topographic data, geological maps, and borehole stratigraphy. The algorithm enhances segmentation by developing a cost function that incorporates topographic roughness, stratigraphic variations, and geological boundary density. This further improves the results by minimizing the cost of inconsistency. It applies dynamic programming to calculate the optimal segmentation points. Through dynamic programming, the model identifies the best segmentation points, ensuring adaptive segmentation. In addition, this paper develops a comprehensive evaluation system for segmentation. The system includes three indicators: the segmentation quality score, modeling efficiency gain, and model accuracy. These indicators allow assessment from three perspectives: the original data used, the benefits for modeling efficiency, and the accuracy of the segmented 3D model. The paper also introduces a formula to measure borehole stratigraphic similarity by combining thickness and category factors. This approach enables a quantitative evaluation of the model’s accuracy. The accuracy is measured by comparing the similarity between the virtual boreholes in the model and the real borehole data. The approach was validated via empirical data from the Xinjiang-Xizang Railway project. The results show that adaptive segmentation reduces the modeling time by approximately 40%, and the model’s accuracy has been improved by at least 5% compared with other segmentation methods. Overall, the proposed framework offers a reliable solution for 3D geological modeling of large-scale linear projects.