A realistic 3D geological model is a vital digital framework in many fields, such as analyzing complex underground structures, conducting engineering surveys, and prospecting mineral resource, etc. However, the accuracy and reliability of 3D geological models are significantly affected by difficulties in data acquisition, data sparsity, and insufficient geological understanding, which limit their in-depth application. Based on the adaptive fully connected deep neural network (AFCDNN) and multi-scale iterative multiple-point statistics (MPS) method, this paper proposes a method for reconstructing 3D geological structures, for which two-dimensional geological cross-sections are used as the modeling data source. Two AFCDNNs are built: one to generate geological surfaces and the other to predict geological attributes. Geological relationships are then rectified in a post-processing manner. An initial model is subsequently constructed. To optimize its local features, the multi-scale iterative MPS method is applied. A concrete example of 3D model construction for a station site of a Guangzhou metro line illustrates that the constructed model correctly reproduces the spatial relationship between faults and strata, achieving a maximum accuracy of 89.3% when validated against actual borehole data. Furthermore, the AFCDNN can generate global features of the geological structure from small sample datasets, and the multi-scale iterative strategy fully leverages the advantages of local optimization. The proposed method exhibits good feasibility, accuracy, and reliability, and can provide an important reference for reconstructing 3D geological structures.
遍历模拟网格,提取已知地质属性区域位置(x, y, z)和相应的地质属性。位置信息采用公式(1)进行归一化处理,地质属性则采用独热编码(One-Hot Encoding)进行特征数字化[48]。
1.4.2 网络构建
属性预测AFCDNN以模拟网格中的空间坐标(x, y, z)为输入,以相应的地质属性为输出,隐藏层之间同样选择ReLU作为激活函数,损失函数采用交叉熵损失函数(公式(3))。采用Adam优化器,通过反向传播算法更新网络权重,以最小化损失函数。与地质面结构预测神经网络不同的是,属性预测深度神经网络在输出层后添加了softmax层,用于将网络输出向量转化为属性预测概率。
LIND, BROEREW, CUIJ Q. Metro systems and urban development: impacts and implications[J]. Tunnelling and Underground Space Technology, 2022, 125: 104509.
[2]
DE RIENZOF, ORESTEP, PELIZZAS. Subsurface geological-geotechnical modelling to sustain underground civil planning[J]. Engineering Geology, 2008, 96(3/4): 187-204.
COMUNIANA, RENARDP, STRAUBHAARJ. 3D multiple-point statistics simulation using 2D training images[J]. Computers & Geosciences, 2012, 40: 49-65.
[10]
LIUG, FANGH F, CHENQ Y, et al. A feature-enhanced MPS approach to reconstruct 3D deposit models using 2D geological cross sections: a case study in the Luodang Cu deposit, southwestern China[J]. Natural Resources Research, 2022, 31(6): 3101-3120.
STREBELLES. Conditional simulation of complex geological structures using multiple-point statistics[J]. Mathematical Geology, 2002, 34: 1-21.
[15]
MARIETHOZG, RENARDP. Reconstruction of incomplete data sets or images using direct sampling[J]. Mathematical Geosciences, 2010, 42: 245-268.
[16]
TAHMASEBIP, HEZARKHANIA, SAHIMIM. Multiple-point geostatistical modeling based on the cross-correlation functions[J]. Computational Geosciences, 2012, 16: 779-797.
[17]
CHENQ Y, MARIETHOZG, LIUG, et al. Locality-based 3-D multiple-point statistics reconstruction using 2-D geological cross sections[J]. Hydrology and Earth System Sciences, 2018, 22(12): 6547-6566.
YANGL, HOUW S, CUIC J, et al. GOSIM: a multi-scale iterative multiple-point statistics algorithm with global optimization[J]. Computers & Geosciences, 2016, 89: 57-70.
[23]
HOUW S, LIUH G, ZHENGT C, et al. Hierarchical MPS-based three-dimensional geological structure reconstruction with two-dimensional image (s)[J]. Journal of Earth Science, 2021, 32(2): 455-467.
[24]
HOUW S, LIUH G, ZHENGT C, et al. Extended GOSIM: MPS-driven simulation of 3D geological structure using 2D cross-sections[J]. Earth and Space Science, 2022, 9(6): e2021EA001801.
[25]
HOUW S, CHENY H, LIUH G, et al. Reconstructing Three-dimensional geological structures by the multiple-point statistics method coupled with a deep neural network: a case study of a metro station in Guangzhou, China[J]. Tunnelling and Underground Space Technology, 2023, 136: 105089.
[26]
ZUOR G, XIONGY H, WANGJ, et al. Deep learning and its application in geochemical mapping[J]. Earth-Science Reviews, 2019, 192: 1-14.
SHAHAMT R, DEKELT, MICHAELIT. Singan: learning a generative model from a single natural image[C]// Proceedings of the IEEE/CVF international conference on Computer Vision. Seoul, Korea (South):2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019: 4570-4580.
[32]
ZHANGT, YANGZ H, SUNC C. Stochastic simulation of fan deltas using parallel multi-stage generative adversarial networks[J]. Journal of Petroleum Science and Engineering, 2022, 208: 109442.
FENGJ X, TENGQ Z, LIB, et al. An end-to-end three-dimensional reconstruction framework of porous media from a single two-dimensional image based on deep learning[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 368: 113043.
[35]
KIMH S, JIY. Three-dimensional geotechnical-layer mapping in Seoul using borehole database and deep neural network-based model[J]. Engineering Geology, 2022, 297: 106489.
[36]
GUOJ T, XUX C, WANGX L, et al. GeoPDNN: a semisupervised deep learning neural network using Pseudolabels for three-dimensional urban geological modelling and uncertainty analysis from borehole data[J]. Geoscientific Model Development Discussions, 2023, 2023: 1-21.
FENGJ X, TENGQ Z, HEX H, et al. Accelerating multi-point statistics reconstruction method for porous media via deep learning[J]. Acta Materialia, 2018, 159: 296-308.
[40]
SCHOLZEG P, BASSANIM A A, COSTAJ F C L. Generative Adversarial Networks to incorporate the training image uncertainty in multiple-point statistics simulation[J]. Geoenergy Science and Engineering, 2023, 230: 212257.
[41]
WANGX X, YUS Y, LIS H, et al. Two parameter optimization methods of multi-point geostatistics[J]. Journal of Petroleum Science and Engineering, 2022, 208: 109724.
MONTAVONG, SAMEKW, MüLLERK R. Methods for interpreting and understanding deep neural networks[J]. Digital Signal Processing, 2018, 73: 1-15.
[45]
BERGSTRAJ, KOMERB, ELIASMITHC, et al. Hyperopt: a python library for model selection and hyperparameter optimization[J]. Computational Science & Discovery, 2015, 8(1): 014008.
[46]
FEURERM, HUTTERF. Hyperparameter optimization[M]. Cham: Springer International Publishing, 2019: 3-33.
[47]
AKIBAT, SANOS, YANASET, et al. Optuna: a next-generation hyperparameter optimization framework[C]// Proceedings of the 25th ACM SIGKDD international conference on Knowledge Discovery & Data Mining. New York: Association for Computing Machinery, 2019: 2623-2631.
[48]
AL-SHEHARIT, ALSOWAILR A. An insider data leakage detection using one-hot encoding, synthetic minority oversampling and machine learning techniques[J]. Entropy, 2021, 23(10): 1258.
SHIC, WANGY. Data-driven construction of three-dimensional subsurface geological models from limited site-specific boreholes and prior geological knowledge for underground digital twin[J]. Tunnelling and Underground Space Technology, 2022, 126: 104493.