姜秀峰，庄培芝，钱远顺，等. 基于稀疏地勘数据的岩土地层信息智能解析研究[J].河南理工大学学报(自然科学版), doi: 10.16186/j.cnki.1673-9787.2024110035.

JIANG X F, ZHUANG P Z, QIAN Y S, et al. Research on intelligent analysis of stratigraphic information based on sparse geological survey data [J]. Journal of Henan Polytechnic University (Natural Science), doi: 10.16186/j.cnki.1673-9787.2024110035.

基于稀疏地勘数据的岩土地层信息智能解析研究

姜秀峰¹，庄培芝²，钱远顺²，厉超吉²，石光³，张明晶³⁴

（1.山东高速建设管理集团有限公司，山东 济南250098；2.山东大学 齐鲁交通学院，山东 济南 250002；3.山东省交通规划设计院集团有限公司，山东 济南 250101）

摘要: 目的 地质勘查中，勘查数据稀疏时，传统数据处理手段和深度学习技术难以准确高效地获取岩土地层空间分布特征，基于此，本文对基于稀疏地勘数据的地层空间分布预测方法进行开发研究。方法 首先使用矩阵方法对已知稀疏地勘数据进行转化，然后采用循环卷积集成算法对参考图进行地层分布特征提取，并使用提取特征对分类器进行训练，随后基于训练的分类器对待测地层图进行多尺度循环插值计算，最终形成二维待测岩土地层空间分布。进一步地，在2张互相垂直的参考地层图基础上，采用本文方法对三维空间的地层分布进行特征提取，并将提取特征与稀疏地勘数据进行循环插值计算，最终实现三维地层模型的预测构建。结果 以实际地层分布图为标准计算预测精度，在使用单张地层参考图和6条稀疏数据的二维验证算例中，本文方法预测准确度为96%；在基于2张地层参考图和36条稀疏地勘数据的三维验算算例中，本文方法预测准确度为92.3%，说明本文方法预测精度高于现行地质勘查处理手段和传统机器学习方法。结论 本文提出的算法可有效提取地层空间分布特征，可仅使用稀疏地勘数据对地层空间分布进行高效准确预测，其预测精度和计算效率满足实际勘察精度要求和工程需求，可进一步扩展到地质灾害防治和矿产开发中。

关键词: 稀疏数据；卷积神经网络；集成学习算法；地层重构

中图分类号：U412.22

doi: 10.16186/j.cnki.1673-9787.2024110035

基金项目: 国家自然科学基金资助项目（52108374）；山东省交通运输厅科技项目（2022B82）

收稿日期：2024-11-19

修回日期：2025-02-24

网络首发日期：2025-10-23

Research on intelligent analysis of stratigraphic information based on sparse geological survey data

Jiang Xiufeng¹,Zhuang Peizhi²,Qian Yuanshun²,Li Chaoji²,Shi Guang³,Zhang Mingjing³

(1. Shandong Hi-Speed Construction Management Group Co., Ltd., Jinan 250098，Shandong，China；2. School of  Qilu Transportation, Shandong University, Jinan 250002，Shandong，China;3. Shandong Provincial Communication Planning and Design Institute Group Co.， Ltd.， Jinan 250101，Shandong，China)

Abstract: Objectives In response to the difficulties in accurately and efficiently obtaining the spatial distribution of stratum using traditional geological exploration data, data processing methods, and conventional deep learning techniques, this study develops a method for predicting the spatial distribution of stratum based on sparse exploration data. Methods First, the known sparse exploration data is transformed using matrix methods. Subsequently, a convolutional integration algorithm is used to extract stratum distribution features from reference maps, and the extracted features are used to train a classifier. Then, based on the trained classifier, multi-scale cyclic interpolation calculations are performed on the stratum maps to be predicted, ultimately forming a two-dimensional spatial distribution of the predicted stratum. Furthermore, based on two reference stratum maps that are perpendicular to each other, the method proposed in this study is used to extract features of the three-dimensional spatial distribution of strata, and the extracted features are cyclically interpolated with the sparse exploration data to ultimately achieve the prediction and construction of a three-dimensional stratum model. Results Taking the actual stratum distribution map as the standard for calculating prediction accuracy, in the two-dimensional verification example using a single stratum reference map and 6 sparse data points, the prediction accuracy of this method is 96%. In the three-dimensional verification example based on 2 stratum reference maps and 36 sparse exploration data points, the prediction accuracy is 92.3%. The prediction accuracy of this method is higher than that of current geological exploration processing methods and traditional machine learning methods. Conclusions The method proposed in this study can effectively extract the spatial distribution features of strata and can predict the spatial distribution of strata efficiently and accurately using only sparse exploration data. The prediction accuracy and computational efficiency meet the actual exploration accuracy requirements and engineering needs. This method can be further extended to geological disaster prevention and mineral development.

Key words：sparse data; convolutional neural network; integrated learning algorithm; stratigraphic reconstruction

CLC: U412.22