A geological profile processing method based on multi-dimensional collaboration of cross-hole CT and borehole information

Abstract

The present invention discloses a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information, and relates to the technical field of intelligent processing of engineering geological image results. The cross-hole CT data of the present invention can provide continuous distribution information of rock and soil bodies between boreholes. Combined with the drilling data, the lithologic distribution can be more finely portrayed. The advantage of cross-hole CT in detecting and identifying the structure of rock and soil bodies between boreholes is combined with the advantage of drilling holes in obtaining the distribution of cores at different depths. The structural characteristics of rock and soil bodies between boreholes obtained by cross-hole CT are utilized and combined with artificial intelligence to realize intelligent correction of engineering geological profiles based on multi-dimensional collaboration of cross-hole CT and drilling information. The corrected engineering geological profiles are more detailed and closer to the actual situation, and have better reference value for engineering design and engineering decision-making. Accurate geological information can reduce the risk of encountering emergencies during construction, reduce the demand for the number of boreholes, and thus reduce exploration costs and time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent processing of engineering geological image results, and in particular to a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information. Background Art

[0002] Drilling refers to the use of specialized drilling equipment to drill into underground rock and soil masses on the surface to form cylindrical holes according to predetermined position, direction, angle and depth requirements. This method can obtain rock cores of rock and soil masses at different depths underground. This is the basis for drawing traditional engineering geological profiles. According to the distribution and related properties of rock cores at different depths, and based on this, reasonable speculation and extension of the areas between boreholes are made, and finally an engineering geological profile is drawn.

[0003] The current method of drawing engineering geological profiles is to extrapolate and extend the areas between boreholes based on the distribution of drill cores. This means that the rock and soil structure between the boreholes is extrapolated rather than the actual rock and soil structure. Especially in limestone strata, the development of caves varies in various forms, and there may be significant differences between the extrapolated rock and soil structure and the actual rock and soil structure.

[0004] The cross-hole CT method is a geophysical exploration method that uses the differences in the propagation characteristics of elastic waves or electromagnetic waves in the medium to perform imaging analysis on the geological body between two boreholes. This method can obtain the absorption characteristics of the rock and soil medium between the boreholes for elastic waves or electromagnetic waves, and then infer the structural characteristics of the rock and soil between the boreholes. However, this method cannot identify the properties of the rock and has certain errors.

[0005] Therefore, it is urgent to design an intelligent correction method for engineering geological profiles that can achieve multi-dimensional collaboration between cross-hole CT and drilling information to solve the technical problem that engineering geological profiles are not detailed enough. Summary of the Invention

[0006] In order to solve the technical problem of intelligent correction of engineering geological profiles based on multi-dimensional collaboration of cross-hole CT and borehole information, the present invention provides a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and borehole information. The following technical solutions are adopted:

[0007] The geological profile processing method based on multi-dimensional collaboration of cross-hole CT and borehole information includes the following steps:

[0008] Step 1, obtaining the graphical results of the cross-hole CT method between drilling holes;

[0009] Step 2: Matrix the graphic results, extract one or more columns of data according to the set horizontal length, and number them according to the horizontal length;

[0010] Step 3, obtaining core histograms of borehole 1 and borehole 2;

[0011] Step 4: Map the No. 0 data of the cross-hole CT data with the depth parameter and lithology parameter of borehole 1, and map the No. 1 data of the cross-hole CT data with the depth parameter and lithology parameter of borehole 2 to obtain mapping regularity data;

[0012] Step 5: perform depth segmentation on the remaining columns of the cross-hole CT data, based on the principle of data mutation, and calculate the average value of each segment;

[0013] Step 6: Apply the up-mapping rule obtained in step 4 to the remaining columns of the cross-hole CT data to convert the cross-hole CT data into corresponding lithology and depth parameters;

[0014] Step 7: Submit the lithology and depth parameters corresponding to all cross-hole CT column data to the trained artificial intelligence model to generate an engineering geological profile.

[0015] By adopting the above technical solution and combining it with cross-hole CT data, the rock and soil structure between boreholes can be reflected more accurately and the estimation error can be reduced.

[0016] Cross-hole CT data can provide continuous distribution information of rock and soil between boreholes. Combined with borehole data, it can more finely characterize the distribution of lithology, such as caves and faults.

[0017] Cross-hole CT data can identify special geological bodies between boreholes that are difficult to obtain through drilling, such as small-scale caves and cracks, providing important reference for engineering design and construction.

[0018] Combining the advantages of cross-hole CT in detecting and identifying the inter-hole rock and soil structure with the advantages of drilling in obtaining the distribution of cores at different depths, and utilizing the inter-hole rock and soil structural characteristics obtained by cross-hole CT and combining them with artificial intelligence, intelligent correction of engineering geological profiles based on multi-dimensional collaboration of cross-hole CT and drilling information is achieved. The corrected engineering geological profiles are more detailed and closer to the actual situation, providing better reference value for engineering design and engineering decision-making.

[0019] Accurate geological information can reduce the risk of unexpected situations during construction, such as caves and faults, thereby ensuring construction safety. It can also reduce the number of holes required, thereby reducing exploration costs and time.

[0020] Optionally, in step 1, a graphical result of implementing the cross-hole CT method between boreholes is obtained by inversion using existing commercial software, and the graphical result is a result map of apparent elastic wave velocity distribution contour lines.

[0021] By adopting the above-mentioned technical solution, and by using a cross-hole CT method to obtain contour maps of the apparent elastic wave velocity distribution, combined with borehole information, this technical solution can significantly improve the accuracy of engineering geological profiles, reduce engineering risks, improve engineering efficiency, save engineering costs, and promote the development of geological exploration technology, thus possessing significant application value. Geophysical inversion is a core technology that uses systematic mathematical modeling to reversely infer the spatial distribution of subsurface physical parameters from surface geophysical field data (including but not limited to gravity, electromagnetic, and seismic fields). Its essence is to solve constrained optimization problems with multi-solution nonlinear ill-conditioned equations. This article discusses a variety of cross-hole CT inversion algorithms, such as the algebraic reconstruction method (ART), the joint iterative method (SIRT), the conjugate gradient method (CGT), and the back projection method (BPT).

[0022] Optionally, the specific method of data matrixing in step 2 is:

[0023] 1. The original grid data is not lost. CT results are often generated by interpolating the original XYZ data to generate grid data, which is then drawn into contour results. If the grid data can be obtained, then the grid data is the data matrix of the result, which contains coordinate points and data information evenly distributed within the cross-section range.

[0024] 2. Import image data and generate data matrix

[0025] The original grid number has been lost, and only the image file is left. You can matrix the image data as follows:

[0026] (1) Scale conversion: If the image contains a scale (such as "1:5000"), determine the conversion relationship between pixels and actual distance based on the scale (for example: 1 pixel = 0.05 meter).

[0027] (2) Automated edge detection for key geological interface extraction: Image algorithms are used to identify continuous curves such as stratum boundaries and fault lines. Adaptive threshold segmentation technology is used for fuzzy boundaries to distinguish different physical property areas, making it easier for spatial sampling in step 4 to pick up physical property information at key points.

[0028] (3) Generate a coordinate system mapping table to establish a two-dimensional coordinate system: Assume that the upper left corner of the image is the origin (0,0), the horizontal direction is the X axis (distance), and the vertical direction is the Y axis (depth / elevation).

[0029] (4) Uniform spatial point sampling: First, grid coverage is performed, and uniform grid points are generated on the profile at a specified spacing (e.g., every 0.1 meter horizontally and every 0.1 meter vertically). A dynamic encryption strategy can be used to automatically increase the sampling density in areas with drastic lithologic changes (e.g., near faults). Then, coordinate inversion and annotation are performed, and each grid point is assigned a real coordinate (X, Y) based on the preset grid spacing and coordinate mapping relationship. For example: a point in the image is located at the 120th pixel (horizontally) × 300 pixels (vertically) → converted to real coordinates (6 meters, 15 meters).

[0030] (5) If the image contains a color code (e.g., color corresponds to wave velocity value), the wave velocity data information is extracted through the color value-parameter comparison table, and finally a CSV / Excel file is generated, which contains three columns of data: X coordinate, Y coordinate, and attribute value (e.g., wave velocity) to obtain a data matrix.

[0031] Optionally, the specific method for obtaining the mapping regularity data in step 4 includes the following steps:

[0032] Step 41: Based on the lithologic distribution depth section of the same stratum in the borehole histogram, find the inversion result value of the corresponding CT result of the borehole in the depth section and calculate the average value to obtain the mapping between the CT result value of the corresponding lithologic type of the borehole and the lithologic type;

[0033] Step 42, sequentially calculating all lithologic parameters and their corresponding CT values in the borehole histogram to obtain a mapping rule for the lithology;

[0034] Step 43: Obtain mapping regularity data of the cross-hole CT data corresponding to the two boreholes.

[0035] By adopting the above technical effects, the same rock layer in the drill hole histogram refers to the rock layers with the same lithology and similar depths in the drill hole histogram. Mainly based on experience, 30% of the hole spacing can be taken as the threshold. If the depth difference is less than this value, it is considered to be the same rock layer.

[0036] Rock layers with the same lithology and similar depth in the drill hole histogram, such as the cave itself and the limestone, sandstone, and overburden (soil layer) above it;

[0037] Mapping rules, based on the direct correspondence between drill core histograms and CT data, eliminate the subjectivity and uncertainty associated with manual lithology judgment and improve the accuracy of lithology identification. Mapping rules can quickly convert CT data into lithology information, enhancing the efficiency of lithology identification. Mapping rules can provide more detailed lithology distribution information, such as lithology variations at different depths and locations, thereby more accurately reflecting the true state of the rock mass between boreholes.

[0038] Mapping rules can help identify and characterize complex geological structures, such as faults, fissures, and caves, providing more comprehensive geological information for engineering design and construction.

[0039] Optionally, the deep segmentation method in step 5 is to use an algorithm capable of identifying data mutation points, and the algorithm capable of identifying data mutation points is a moving average method or a first-order difference method.

[0040] By adopting the above technical solution, the extracted column data is deeply segmented by using an algorithm that has the ability to identify data mutation points. The algorithm that has the ability to identify data mutation points is the moving average method or the first-order difference method. This method can identify the numerical mutation characteristics to segment the column data, and then calculate the average value of each segment after segmentation to obtain the lithologic parameters in reverse combination with the mapping relationship.

[0041] Optionally, the artificial intelligence model in step 7 is based on a graph neural network architecture.

[0042] By adopting the above technical solutions, the artificial intelligence model architecture is designed as follows:

[0043] Input layer: 1.1 Data source, structured table data: location (X), depth range (Zmin, Zmax), physical properties (wave velocity, electromagnetic absorption coefficient, etc.), lithology label. Contains multiple data points, one for each location X.

[0044] 1.2 Preprocessing, Depth Range Parsing: Convert a depth range in string format (e.g., "10-20") into a numeric pair (Z_top, Z_bottom). Feature Normalization: Normalize physical property parameters (e.g., Min-Max or Z-Score). Spatial Encoding: Map position coordinates (X, 0) to nodes in a grid or graph structure.

[0045] Core model layer:

[0046] 2.1 Lithologic boundary prediction module:

[0047] 1) Model selection:

[0048] Graph Neural Network (GNN): Models the spatial dependencies between data points (e.g., lithologic continuity of adjacent lines). The neighborhood aggregation formula is as follows:

[0049]

[0050] vi: The feature vector of data point i, including its location (Xi), depth range (Ztop,i, Zbottom,i), and physical property parameters. N(i): The set of spatial neighbors of data point i. hi: The hidden representation of data point i after GNN aggregation, which is the fusion of spatial features and physical property parameters. It transforms discrete data points into continuous geological structure information, providing key input for subsequent lithologic classification and profile generation. This process simulates the geologist's way of thinking, "inferring strata based on surrounding drill holes," but is automated through data-driven automation.

[0051] Spatial dependency modeling: Capture the spatial correlation between data points by aggregating the information of neighboring nodes {vj}. For example, the lithology between adjacent data points on the same survey line is usually continuous (such as a gradual transition from sandstone to shale). The aggregation method is as follows:

[0052] Mean Pooling:

[0053] Attention Mechanism:

[0054] W: trainable weight matrix.

[0055] If the physical properties of data point i are abnormally high, but its surrounding neighbors are all sandstone, GNN will reduce the probability of the point being misclassified as limestone.

[0056] Bidirectional LSTM / Transformer: Captures vertical changes in lithology along the depth direction (Z-axis).

[0057]

[0058] Zi: The depth value of data point i (e.g., Ztop, i or midpoint depth). sz: The hidden state of the depth sequence output by the LSTM, encoding the lithologic variation pattern along the depth direction. Vertical sequence modeling: The LSTM processes data points in depth order (from top to bottom or bottom to top) to capture the vertical regularity of lithologic properties.

[0059] 2) Input: Feature vector of each data point [X, Z_top, Z_bottom, physical properties].

[0060] Output: lithology label probability distribution P(L|X,Z).

[0061] Probabilistic classification: Map the hidden state sz output by the LSTM to a probability distribution of lithology labels. Multi-classification support: Softmax ensures that the sum of all lithology probabilities is 1.

[0062] ;

[0063] W: trainable weight matrix. b: bias vector. P(Li=l): probability that data point i belongs to lithology l.

[0064] Through the above process, the model can reconstruct a continuous and reasonable geological profile from discrete data points, which is significantly better than traditional threshold segmentation or interpolation methods.

[0065] 2.2 Lithologic regional continuity modeling module:

[0066] The goal is to connect the discrete prediction results into continuous lithologic areas and resolve fault and noise issues.

[0067] Method,Conditional Random Field (CRF): Constrain the lithology consistency of adjacent nodes through energy function.Morphological post-processing: Use dilation / erosion operations to smooth the boundaries.

[0068] Energy function:

[0069]

[0070] in: is the single-point lithology probability (from GNN-LSTM). Penalizes the sudden change of lithology between adjacent points (i, j).

[0071] Profile generation layer:

[0072] To complete polygon construction and filling, we first extract the top and bottom boundaries: for each lithology label, we extract the continuous top boundary (Z_top) and bottom boundary (Z_bottom) along the survey line position (X). Next, we generate a closed region: we connect the top and bottom boundary points to form a polygon vertex sequence. We then perform anti-aliasing: we use Bezier curves to smooth the boundaries. Finally, we generate a unique fill color or pattern based on the lithology hash value (e.g., sandstone → yellow, shale → gray).

[0073] Training and optimization strategies:

[0074] Loss function:

[0075]

[0076] Classification loss: Cross-Entropy loss optimizes lithology prediction:

[0077] Boundary smoothing loss: CRF energy term constrains spatial continuity

[0078] Distributed training: Use Horovod or PyTorchDistributed to accelerate large-scale test data training.

[0079] Advantages:

[0080] Multimodal fusion, while utilizing physical parameters and spatial topological relationships, improves the prediction accuracy of complex geological structures.

[0081] End-to-end automation: no manual intervention is required for the entire process from raw tabular data to visual profiles.

[0082] Dynamic adaptability, through CRF and morphological processing, is compatible with heterogeneous geological phenomena such as faults and pinch-outs.

[0083] A storage medium is used to store a processing program designed using a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information.

[0084] A computer runs a processing program designed based on a geological profile processing method of multi-dimensional collaboration between cross-hole CT and borehole information, and outputs an intelligently corrected engineering geological profile.

[0085] In summary, the present invention includes at least one of the following beneficial technical effects:

[0086] The present invention can provide a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and borehole information. Cross-hole CT data can provide continuous distribution information of rock and soil between boreholes. Combined with borehole data, it can more finely depict the lithologic distribution.

[0087] Cross-hole CT data can identify special geological bodies between boreholes that are difficult to access through drilling;

[0088] Combining the advantages of cross-hole CT in detecting and identifying the inter-hole rock and soil structure with the advantages of drilling in obtaining the distribution of cores at different depths, and utilizing the inter-hole rock and soil structural characteristics obtained by cross-hole CT and combining them with artificial intelligence, intelligent correction of engineering geological profiles based on multi-dimensional collaboration of cross-hole CT and drilling information is achieved. The corrected engineering geological profiles are more detailed and closer to the actual situation, providing better reference value for engineering design and engineering decision-making.

[0089] Accurate geological information can reduce the risk of unexpected situations during construction, such as caves and faults, thereby ensuring construction safety. It can also reduce the number of holes required, thereby reducing exploration costs and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a flow chart of a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information of the present invention;

[0091] FIG2( a ) is a histogram of the drilling of the drilling hole ZK01 in a specific embodiment of the present invention;

[0092] FIG2( b ) is a histogram of the drilling of the drilling hole ZK02 in a specific embodiment of the present invention;

[0093] Figure 3 This is a schematic diagram of a traditional engineering geological section in a specific embodiment of the present invention;

[0094] Figure 4 This is a schematic diagram of cross-hole seismic wave CT graphics results in a specific embodiment of the present invention;

[0095] Figure 5 2 is a schematic diagram of a numerical matrix of cross-hole CT results in a specific embodiment of the present invention;

[0096] Figure 6 It is a schematic diagram of the corrected engineering geological profile in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0097] The present invention will be further described in detail below with reference to the accompanying drawings.

[0098] The embodiment of the present invention discloses a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information.

[0099] Reference Figures 1-6 , Example 1, a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information, comprising the following steps:

[0100] Step 1, obtaining the graphical results of the cross-hole CT method between drilling holes;

[0101] Step 2: Matrix the graphic results, extract one or more columns of data according to the set horizontal length, and number them according to the horizontal length;

[0102] Step 3, obtaining core histograms of borehole 1 and borehole 2;

[0103] Step 4: Map the No. 0 data of the cross-hole CT data with the depth parameter and lithology parameter of borehole 1, and map the No. 1 data of the cross-hole CT data with the depth parameter and lithology parameter of borehole 2 to obtain mapping regularity data;

[0104] Step 5: perform depth segmentation on the remaining columns of the cross-hole CT data, based on the principle of data mutation, and calculate the average value of each segment;

[0105] Step 6: Apply the up-mapping rule obtained in step 4 to the remaining columns of the cross-hole CT data to convert the cross-hole CT data into corresponding lithology and depth parameters;

[0106] Step 7: Submit the lithology and depth parameters corresponding to all cross-hole CT column data to the trained artificial intelligence model to generate an engineering geological profile.

[0107] Combined with cross-hole CT data, it can more accurately reflect the rock and soil structure between boreholes and reduce inference errors.

[0108] Cross-hole CT data can provide continuous distribution information of rock and soil between boreholes. Combined with borehole data, it can more finely characterize the distribution of lithology, such as caves and faults.

[0109] Cross-hole CT data can identify special geological bodies between boreholes that are difficult to obtain through drilling, such as small-scale caves and cracks, providing important reference for engineering design and construction.

[0110] Combining the advantages of cross-hole CT in detecting and identifying the inter-hole rock and soil structure with the advantages of drilling in obtaining the distribution of cores at different depths, and utilizing the inter-hole rock and soil structural characteristics obtained by cross-hole CT and combining them with artificial intelligence, intelligent correction of engineering geological profiles based on multi-dimensional collaboration of cross-hole CT and drilling information is achieved. The corrected engineering geological profiles are more detailed and closer to the actual situation, providing better reference value for engineering design and engineering decision-making.

[0111] Accurate geological information can reduce the risk of unexpected situations during construction, such as caves and faults, thereby ensuring construction safety. It can also reduce the number of holes required, thereby reducing exploration costs and time.

[0112] In Example 2, in step 1, the graphical results of implementing the cross-hole CT method between boreholes are obtained by inversion using existing commercial software. The graphical results are contour lines of apparent elastic wave velocity distribution.

[0113] By using the cross-hole CT method to obtain the apparent elastic wave velocity distribution contour map and combining it with drilling information, this technical solution can significantly improve the accuracy of engineering geological profiles, reduce engineering risks, improve engineering efficiency, save engineering costs, and promote the development of geological exploration technology, with important application value.

[0114] Geophysical inversion is a core technology that uses systematic mathematical modeling to infer the spatial distribution of subsurface physical properties from surface geophysical field data (including but not limited to gravity, electromagnetic, and seismic fields). Its essence lies in solving constrained optimization problems involving multi-solution nonlinear ill-conditioned equations. This article discusses various cross-hole CT inversion algorithms, including the algebraic reconstruction method (ART), the joint iterative method (SIRT), the conjugate gradient method (CGT), and the back projection technique (BPT).

[0115] Geophysical forward modeling is the simulation of geophysical field responses (such as gravity anomalies, seismic wave fields, and electromagnetic signals) observed on the surface or in wells, based on known subsurface physical parameters (such as density, velocity, and resistivity) using physical laws and mathematical models. It is one of the core tools in geophysical exploration and research, providing a theoretical basis for data interpretation, inversion, and geological modeling.

[0116] Taking the conjugate gradient method (CGT) as an example, its specific implementation steps are as follows:

[0117] Data preparation: Collect cross-hole electromagnetic wave observation data, including the locations of the transmitting and receiving points, the travel time and amplitude of the electromagnetic waves, and other information.

[0118] Initial model setting: Set the initial parameter distribution of the subsurface medium, such as a uniform model or a model based on prior information.

[0119] Forward modeling: Use the current model to calculate predicted data, such as travel time or attenuation for each ray.

[0120] Calculate the residual: the difference Δd between the observed data and the predicted data.

[0121] Construct the Jacobian matrix J: Calculate the sensitivity of each model parameter to each ray data.

[0122] The inversion equation is constructed, and the regularization term is considered to form a linear system.

[0123] The conjugate gradient method is used to solve the model update Δm.

[0124] Update the model: m = m + Δm.

[0125] Determine convergence: Check whether the residual is small enough or whether the number of iterations reaches the preset value. If not, return to step 3.

[0126] Output the final inversion results.

[0127] The above steps can be implemented with mature commercial software, such as the Electromagnetic Wave CT Multi-hole Joint Inversion Software V2.1 and the Electromagnetic Wave CT Inversion Software V2.6.

[0128] In Example 3, the specific method of data matrixing in step 2 is:

[0129] 1. The original grid data is not lost. CT results are often generated by interpolating the original XYZ data to generate grid data, which is then drawn into contour results. If the grid data can be obtained, then the grid data is the data matrix of the result, which contains coordinate points and data information evenly distributed within the cross-section range.

[0130] 2. Import image data and generate data matrix

[0131] The original grid number has been lost, and only the image file is left. You can matrix the image data as follows:

[0132] (1) Scale conversion: If the image contains a scale (such as "1:5000"), determine the conversion relationship between pixels and actual distance based on the scale (for example: 1 pixel = 0.05 meter).

[0133] (2) Automated edge detection for key geological interface extraction: Image algorithms are used to identify continuous curves such as stratum boundaries and fault lines. Adaptive threshold segmentation technology is used for fuzzy boundaries to distinguish different physical property areas, making it easier for spatial sampling in step 4 to pick up physical property information at key points.

[0134] (3) Generate a coordinate system mapping table to establish a two-dimensional coordinate system: Assume that the upper left corner of the image is the origin (0,0), the horizontal direction is the X axis (distance), and the vertical direction is the Y axis (depth / elevation).

[0135] (4) Uniform spatial point sampling: First, grid coverage is performed, and uniform grid points are generated on the profile at a specified spacing (e.g., every 0.1 meter horizontally and every 0.1 meter vertically). A dynamic encryption strategy can be used to automatically increase the sampling density in areas with drastic lithologic changes (e.g., near faults). Then, coordinate inversion and annotation are performed, and each grid point is assigned a real coordinate (X, Y) based on the preset grid spacing and coordinate mapping relationship. For example: a point in the image is located at the 120th pixel (horizontally) × 300 pixels (vertically) → converted to real coordinates (6 meters, 15 meters).

[0136] (5) If the image contains a color code (e.g., color corresponds to wave velocity value), the wave velocity data information is extracted through the color value-parameter comparison table, and finally a CSV / Excel file is generated, which contains three columns of data: X coordinate, Y coordinate, and attribute value (e.g., wave velocity) to obtain a data matrix.

[0137] In Example 4, the specific method for obtaining mapping regularity data in step 4 includes the following steps:

[0138] Step 41: Based on the lithologic distribution depth section of the same stratum in the borehole histogram, find the inversion result value of the CT result corresponding to the borehole in the depth section and calculate the average value to obtain the mapping between the CT result value corresponding to the borehole and the lithologic property;

[0139] Step 42, sequentially calculating all lithologic parameters and their corresponding CT values in the borehole histogram to obtain a mapping law of lithologic properties;

[0140] Step 43: Obtain mapping regularity data of the cross-hole CT data corresponding to the two boreholes.

[0141] The same rock layer in the drill hole histogram refers to rock layers with the same lithology and similar depths in the drill hole histogram. Mainly based on experience, 30% of the hole spacing can be taken as the threshold. If the depth difference is less than this value, it is considered to be the same rock layer.

[0142] Mapping rules, based on the direct correspondence between drill core histograms and CT data, eliminate the subjectivity and uncertainty associated with manual lithology judgment and improve the accuracy of lithology identification. Mapping rules can quickly convert CT data into lithology information, enhancing the efficiency of lithology identification. Mapping rules can provide more detailed lithology distribution information, such as lithology variations at different depths and locations, thereby more accurately reflecting the true state of the rock mass between boreholes.

[0143] Mapping rules can help identify and characterize complex geological structures, such as faults, fissures, and caves, providing more comprehensive geological information for engineering design and construction.

[0144] In Example 5, the deep segmentation method in step 5 adopts an algorithm capable of identifying data mutation points, and the algorithm capable of identifying data mutation points is a moving average method or a first-order difference method.

[0145] The extracted column data is deeply segmented by using an algorithm that has the ability to identify data mutation points. The algorithm that has the ability to identify data mutation points is the moving average method or the first-order difference method. This method can identify the numerical mutation characteristics to segment the column data, and then calculate the average value of each segment after segmentation to obtain the lithologic parameters in reverse by combining the mapping relationship.

[0146] In Example 6, the artificial intelligence model in step 7 is based on a graph neural network architecture.

[0147] The artificial intelligence model architecture is designed as follows:

[0148] Input layer: 1.1 Data source, structured table data: location (X), depth range (Zmin, Zmax), physical properties (wave velocity, electromagnetic absorption coefficient, etc.), lithology label. Contains multiple data points, one for each location X.

[0149] 1.2 Preprocessing, Depth Range Parsing: Convert a depth range in string format (e.g., "10-20") into a numeric pair (Z_top, Z_bottom). Feature Normalization: Normalize physical property parameters (e.g., Min-Max or Z-Score). Spatial Encoding: Map position coordinates (X, 0) to nodes in a grid or graph structure.

[0150] Core model layer:

[0151] 2.1 Lithologic boundary prediction module:

[0152] 1) Model selection:

[0153] Graph Neural Network (GNN): Models the spatial dependencies between data points (e.g., lithologic continuity of adjacent lines). The neighborhood aggregation formula is as follows:

[0154]

[0155] vi: The feature vector of data point i, including its location (Xi), depth range (Ztop,i, Zbottom,i), and physical property parameters. N(i): The set of spatial neighbors of data point i. hi: The hidden representation of data point i after GNN aggregation, which is the fusion of spatial features and physical property parameters. It transforms discrete data points into continuous geological structure information, providing key input for subsequent lithologic classification and profile generation. This process simulates the geologist's way of thinking, "inferring strata based on surrounding drill holes," but is automated through data-driven automation.

[0156] Spatial dependency modeling: Capture the spatial correlation between data points by aggregating the information of neighboring nodes {vj}. For example, the lithology between adjacent data points on the same survey line is usually continuous (such as a gradual transition from sandstone to shale). The aggregation method is as follows:

[0157] Mean Pooling:

[0158] Attention Mechanism:

[0159] W: trainable weight matrix.

[0160] If the physical properties of data point i are abnormally high, but its surrounding neighbors are all sandstone, GNN will reduce the probability of the point being misclassified as limestone.

[0161] Bidirectional LSTM / Transformer: Captures vertical changes in lithology along the depth direction (Z-axis).

[0162]

[0163] Zi: The depth value of data point i (e.g., Ztop, i or midpoint depth). sz: The hidden state of the depth sequence output by the LSTM, encoding the lithologic variation pattern along the depth direction. Vertical sequence modeling: The LSTM processes data points in depth order (from top to bottom or bottom to top) to capture the vertical regularity of lithologic properties.

[0164] 2) Input: Feature vector of each data point [X, Z_top, Z_bottom, physical properties].

[0165] Output: lithology label probability distribution P(L|X,Z).

[0166] Probabilistic classification: Map the hidden state sz output by the LSTM to a probability distribution of lithology labels. Multi-classification support: Softmax ensures that the sum of all lithology probabilities is 1.

[0167] ;

[0168] W: trainable weight matrix. b: bias vector. P(Li=l): probability that data point i belongs to lithology l.

[0169] Through the above process, the model can reconstruct a continuous and reasonable geological profile from discrete data points, which is significantly better than traditional threshold segmentation or interpolation methods.

[0170] 2.2 Lithologic regional continuity modeling module:

[0171] The goal is to connect the discrete prediction results into continuous lithologic areas and resolve fault and noise issues.

[0172] Method,Conditional Random Field (CRF): Constrain the lithology consistency of adjacent nodes through energy function.Morphological post-processing: Use dilation / erosion operations to smooth the boundaries.

[0173] Energy function:

[0174]

[0175] in: is the single-point lithology probability (from GNN-LSTM). Penalizes the sudden change of lithology between adjacent points (i, j).

[0176] Profile generation layer:

[0177] To complete polygon construction and filling, we first extract the top and bottom boundaries: for each lithology label, we extract the continuous top boundary (Z_top) and bottom boundary (Z_bottom) along the survey line position (X). Next, we generate a closed region: we connect the top and bottom boundary points to form a polygon vertex sequence. We then perform anti-aliasing: we use Bezier curves to smooth the boundaries. Finally, we generate a unique fill color or pattern based on the lithology hash value (e.g., sandstone → yellow, shale → gray).

[0178] Training and optimization strategies:

[0179] Loss function:

[0180]

[0181] Classification loss: Cross-Entropy loss optimizes lithology prediction:

[0182] Boundary smoothing loss: CRF energy term constrains spatial continuity

[0183] Distributed training: Use Horovod or PyTorchDistributed to accelerate large-scale test data training.

[0184] Advantages:

[0185] Multimodal fusion, while utilizing physical parameters and spatial topological relationships, improves the prediction accuracy of complex geological structures.

[0186] End-to-end automation: no manual intervention is required for the entire process from raw tabular data to visual profiles.

[0187] Dynamic adaptability, through CRF and morphological processing, is compatible with heterogeneous geological phenomena such as faults and pinch-outs.

[0188] Example 7, a storage medium for storing a processing program designed using a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information.

[0189] Example 8, a computer, which runs a processing program designed using a geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information, and outputs an intelligently corrected engineering geological profile.

[0190] The following uses a specific embodiment to illustrate the implementation principle of the geological profile processing method based on the multi-dimensional collaboration of cross-hole CT and borehole information of the present invention:

[0191] There are two limestone boreholes, ZK01 and ZK02. The histograms are shown in Figures 2(a) and 2(b). The distance between the two boreholes is 5.4 m. The cross-hole seismic wave CT method was implemented between the boreholes. The detection target position was from an elevation of -20 m to the bottom of the borehole. The inversion obtained the contour line results of the apparent elastic wave velocity distribution of the rock and soil structure between the boreholes. The graphical results are shown in Figure 2(a) and (b). Figure 4 .

[0192] Without combining cross-hole CT results, the traditional engineering geological profile under drilling conditions is shown. Figure 3 :

[0193] The cross-hole elastic wave CT results were numerically matrixed, see Figure 5, taking a fixed horizontal length of 1m, a horizontal length of 0m corresponds to borehole ZK01, and a horizontal length of 5.4m corresponds to borehole ZK02. As shown in the ZK01 histogram, the lithology from -20m to -25.51m is semi-rocky and semi-soily strongly weathered mudstone, and the corresponding elastic wave CT results for the range from -20m to 25.51m have an average value of 2522m / s. The lithology from -25.51m to -29.01m is extremely fractured moderately weathered limestone, and the corresponding elastic wave CT results for the range from -25.51m to -29.01m have an average value of 3275m / s. The lithology from -01m to -31.51m is a fully filled cave, and the filling is silty clay with a small amount of limestone dissolution fragments. The corresponding elastic wave CT results from -29.01m to -31.51m have an average data value of 2728m / s. The lithology from -31.51m to -37.51m is a relatively broken medium-weathered limestone. The corresponding elastic wave CT results from -31.51m to -37.51m have an average data value of 3320m / s. As shown in the ZK02 bar chart, the lithology from -20m to -26.71m is semi-rock and semi-soil strongly weathered mudstone, and the corresponding elastic wave CT result data average value of -20m to -26.71m is 2586m / s. The lithology from -26.71m to -27.81m is extremely broken moderately weathered limestone, and the corresponding elastic wave CT result data average value of -26.71m to -27.81m is 3501m / s. The lithology from -27.81m to -37.21m is beaded The lithology of the karst cave is characterized by silty clay and some gravel. The average elastic wave CT data from -27.81 m to -37.21 m is 2858 m / s. The lithology from -37.21 m to -43.21 m is relatively broken and moderately weathered limestone, and the average elastic wave CT data from -37.21 m to -43.21 m is 3386 m / s. The mapping patterns are summarized in Table 1. Due to the close distance between the boreholes, it can be assumed that the mapping patterns between the lithology and CT data are similar. Depth and apparent velocity data corresponding to the elastic wave CT data were extracted with a unit length of 1 m. Five data sets with lateral lengths of 1 m, 2 m, 3 m, 4 m, and 5 m were used to identify the apparent velocity mutation points of these five data sets using the moving average method. The data were segmented according to the depth of the mutation points, and the average apparent velocity of the elastic wave CT results for the corresponding depth segments was extracted. Lithology mapping was then performed based on the parameter mapping pattern between borehole lithology and CT apparent velocity. Detailed parameters are shown in Table 2.

[0194] Table 1 Mapping table of lithology and cross-hole elastic wave CT data

[0195]

[0196] Table 2 Lithological parameters of cross-hole elastic wave CT data mapping

[0197]

[0198] At this point, the lithologic information of the cross-section between boreholes with a spacing of 1m is obtained. This information is submitted to the trained artificial intelligence model to achieve a more refined engineering geological cross-section diagram. The result is shown in the figure below. Figure 6 shown.

[0199] The revised engineering geological profile is significantly different from the traditional engineering geological profile in terms of the performance of caves between boreholes. The revised engineering geological profile is more detailed and closer to the actual situation, and has better reference value for engineering design and engineering decision-making.

[0200] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A geological profile processing method based on multi-dimensional collaboration of cross-hole CT and drilling information, characterized in that: The following steps are involved: Step 1, obtaining the graphical results of the cross-hole CT method between drilling holes; Step 2: Matrix the graphic results, extract one or more columns of data according to the set horizontal length, and number them according to the horizontal length; Step 3, obtaining core histograms of borehole 1 and borehole 2; Step 4: Map the No. 0 data of the cross-hole CT data with the depth parameter and lithology parameter of borehole 1, and map the No. 1 data of the cross-hole CT data with the depth parameter and lithology parameter of borehole 2 to obtain mapping regularity data; Step 5: perform depth segmentation on the remaining columns of the cross-hole CT data, based on the principle of data mutation, and calculate the average value of each segment; Step 6: Apply the up-mapping rule obtained in step 4 to the remaining columns of the cross-hole CT data to convert the cross-hole CT data into corresponding lithology and depth parameters; Step 7: Submit the lithology and depth parameters corresponding to all cross-hole CT column data to the trained artificial intelligence model to generate an engineering geological profile.

2. The geological profile processing method based on cross-hole CT and multi-dimensional collaboration of borehole information according to claim 1 is characterized by: In step 1, the graphical results of the cross-hole CT method implemented between boreholes are obtained by inversion using existing commercial software. The graphical results are contour lines of elastic wave velocity distribution.

3. The geological profile processing method based on cross-hole CT and multi-dimensional collaboration of drilling information according to claim 1 is characterized in that: The specific method for obtaining the mapping regularity data in step 4 includes the following steps: Step 41: Based on the lithologic distribution depth section of the same stratum in the borehole histogram, find the inversion result value of the corresponding CT result of the borehole in the depth section and calculate the average value to obtain the mapping between the CT result value of the corresponding lithologic type of the borehole and the lithologic type; Step 42, sequentially calculating all lithologic parameters and their corresponding CT values in the borehole histogram to obtain a mapping rule for the lithology; Step 43: Obtain mapping regularity data of the cross-hole CT data corresponding to the two boreholes.

4. The geological profile processing method based on cross-hole CT and multi-dimensional collaboration of drilling information according to claim 1 is characterized in that: The deep segmentation method in step 5 is to use an algorithm that has the ability to identify data mutation points. The algorithm that has the ability to identify data mutation points is a moving average method or a first-order difference method.

5. The geological profile processing method based on cross-hole CT and multi-dimensional collaboration of drilling information according to claim 1 is characterized in that: The artificial intelligence model in step 7 is based on the graph neural network architecture.

6. A storage medium, characterized in that Used to store a processing program designed using the geological profile processing method based on cross-hole CT and multi-dimensional collaboration of drilling information as described in any one of claims 1-5.

Images