Multi-point statistics three-dimensional stratum modeling method based on adaptive search

Abstract

The invention provides a multi-point statistics three-dimensional stratum modeling method based on adaptive search, and relates to the field of three-dimensional stratum modeling, and the method comprises the steps: setting a horizontal search range, and obtaining a vertical precise search region through space projection; and taking an intersection of the horizontal search range and the vertical search range as a training image of the current node. And two kinds of aggregation methods are adopted for different training images: an addition aggregation operator is adopted for a parallel section to enlarge the number of training samples, and a multiplication aggregation operator is adopted for an orthogonal section to reserve anisotropic characteristics. In the aggregation process, the area of a training image and the distance between a section and a to-be-simulated point serve as two weight factors, and therefore the control effect of spatial correlation on simulation is embodied. And through probability distribution aggregation of the training image, obtaining a simulation attribute value of the to-be-simulated point from an aggregation function according to a probability random value. And replacing a complete three-dimensional training image with the cross section to realize automatic reconstruction of a complex three-dimensional geological structure model.

Description

Technical Field

[0001] The present application relates to the field of three-dimensional stratum modeling, and in particular to a multi-point statistical three-dimensional stratum modeling method based on adaptive search. Background Art

[0002] 3D structural modeling typically relies on boreholes and profiles as fundamental data sources. As first-hand information, borehole data contains the most reliable topological relationships and attribute information. Most geological profiles are constructed from multiple boreholes, significantly expanding structural understanding at a relatively low cost.

[0003] A few sections come directly from geological outcrops. The most common modeling process is to draw a section with a drill hole, and then use the two-dimensional section to build a three-dimensional model. Three-dimensional structural models are mainly divided into two categories according to their morphology: surface models and volume models. The surface model is a three-dimensional geological model constructed layer by layer based on the spatial topological relationship. It can well reflect the contact relationship and the model is lightweight, but the surface model is difficult to express the feature distribution and underground heterogeneity. The volume model uses a fine grid to depict geological entities. It can not only represent the macro structure, but also describe the continuous change of attributes. Due to its rich information, the volume model is often used for quantitative analysis and engineering decision-making, but its calculation and storage memory overhead is relatively large.

[0004] Based on different modeling function theories, three-dimensional modeling methods can be divided into deterministic modeling and stochastic modeling. Deterministic modeling provides deterministic predictions for unknown areas between wells. It uses first-hand exploration data to infer unique parameters and includes methods such as the inverse distance weighting method, the nearest neighbor method, and the Kriging method. The Kriging method and its variants are the most widely used deterministic modeling methods. Deterministic modeling methods have difficulty describing the heterogeneity of underground structures and are not suitable for uncertainty assessment in engineering geology. Stochastic modeling methods, based on geostatistics and random function theory, provide multiple possible, equally probabilistic predictions for the distribution and variation of geological attribute parameters. Stochastic modeling methods are suitable for automated simulation and stochastic expression, and include two-point statistical theory and multi-point statistical theory. Due to the difficulty in incorporating geological knowledge and constraints, the topological relationships and macrostructures of geological entities are not easily reproduced in stochastic simulations.

[0005] Multi-point geostatistics, a representative example of stochastic modeling theory, aims to reproduce geological heterogeneity by capturing high-order features from training images. It overcomes the shortcomings of variogram-based methods in characterizing complex geological structures and reproducing connectivity. The main advantage of multi-point geostatistics lies in its ability to capture complex geological patterns and express stochasticity. Multi-point statistical stochastic simulations are typically based on a regular grid. A data template within the grid is used to extract a multi-point pattern from the training image. This multi-point pattern consists of multiple points in different orientations within the simulation grid. The training image is a representative image that contains specific heterogeneous patterns of geological structure. The prior patterns and probability distributions of the variables in the training image are reproduced in the target model. The quality of the training image directly determines the quality of the simulation results. After significant development, modeling methods based on multi-point geostatistics have been proposed and applied to the modeling and characterization of structures and properties in various fields, such as mineral resources, oil and gas, and hydrogeology.

[0006] The first effective implementation of multi-point geostatistical stochastic simulation theory employed a global scan of training images during point-by-point simulation. The construction of a search tree enabled the pattern probability distribution of the training images to be extracted and stored all at once, improving simulation efficiency. Furthermore, by introducing a pattern distance to measure the similarity between patterns in the training image and the current point to be simulated, and directly capturing similar geological patterns from the training images, simulation efficiency was significantly improved while ensuring good simulation results. To increase the speed of large-scale simulations, a multi-point statistical hybrid parallel framework was applied to supercomputing platforms. Although numerous algorithms based on multi-point statistical theory exist, most are difficult to apply to practical 3D modeling because they require 3D training images to provide spatial patterns for 3D model reconstruction. Obtaining realistic prior 3D models with rich patterns is extremely difficult, and the lack of 3D training images has limited the widespread use of multi-point geostatistical modeling methods. Compared to 3D training images, boreholes and geological profiles are more readily available. Using 2D profiles as training images and conditioning data to generate 3D geological structural models is undoubtedly a key breakthrough in multi-point geostatistical modeling.

[0007] Multi-point geostatistical theory assumes stationarity, meaning that heterogeneous patterns recur and are evenly distributed throughout the training image. Most multi-point geostatistical stochastic simulation algorithms are designed for geological modeling scenarios with good stationarity. However, stationary data is rarely available in complex geological settings. Sedimentary structures and weathering and erosion form strata, which are arranged vertically over geological time periods, resulting in a non-stationary stratigraphic sequence. Folds, formed by geostress, also exhibit macroscopically non-repeatable and non-stationary morphology. Faults are typically planar, with abrupt lithology changes on either side, resulting in unconformable strata. Complex secondary structures such as lenses, intrusions, and pinchouts make non-stationarity the norm in real geological environments, while stationarity only exists at small scales or microscopic scales. Therefore, even with cross-sectional data, existing multi-point geostatistical stochastic simulation methods struggle to accurately reconstruct macroscopic structures in non-stationary environments. Breaking the constraints of the stationarity assumption is a key issue in multi-point geostatistical stochastic simulation theory.

[0008] In summary, the lack of a rational pattern extraction strategy and the widespread and profound presence of geological nonstationarity make the application of 2D training data to construct 3D geological structural models in complex geological settings extremely difficult. Accordingly, the rational extraction and aggregation of 2D profile data can accurately reconstruct the 3D geological model. Based on the principle of spatial correlation, different simulation points utilize different parts of the profile and assign appropriate aggregation weights to different sections, thereby distinguishing and unifying local stationarity and global nonstationarity. Given the above analysis and the problems and challenges faced by multi-point geostatistical modeling methods in practical applications, it is necessary to propose a multi-point statistical 3D stratigraphic modeling method based on adaptive search. Summary of the Invention

[0009] The purpose of the present invention is to provide a multi-point statistical three-dimensional stratum modeling method based on adaptive search in order to solve the problems of prominent geological heterogeneity and complex stratigraphic topology in existing three-dimensional stratum modeling methods.

[0010] The above-mentioned purpose of this application is achieved through the following technical solutions:

[0011] S1: Acquire two-dimensional geological data and assign data points of the two-dimensional geological data to a simulation grid;

[0012] S2: Set the parameters of the simulation grid and the simulation path;

[0013] S3: Traverse the simulation path in a preset order to obtain the current simulation point P; if it is determined that the traversal is not completed, jump to step S4; if it is determined that the traversal is completed, jump to step S8;

[0014] S4: Obtain the coordinates of the current simulation point P and the neighbor profile closest to the current simulation point P;

[0015] S5: traverse the neighbor profile to obtain the probability distribution function of the training image;

[0016] S6: Perform probability aggregation on the probability distribution function to obtain a joint probability distribution function;

[0017] S7: Determine the simulation attribute value of the current simulation point P through the joint probability distribution function; jump to step S3;

[0018] S8: Save the current simulation results and complete the three-dimensional reconstruction of all unknown positions on the simulation grid.

[0019] Optionally, step S1 includes:

[0020] The two-dimensional geological data includes: horizontal and vertical geological profile data, remote sensing images and drilling sampling data.

[0021] Optionally, step S2 includes:

[0022] The parameters of the simulation grid include: the size of the simulation grid and the size of the multi-point statistical template T;

[0023] The simulation path covers all unknown nodes on the current simulation grid.

[0024] Optionally, step S4 includes:

[0025] The nearest section includes: a neighbor section closest to the point P in the X-axis direction or a neighbor section closest to the point P in the Y-axis direction.

[0026] Optionally, step S5 includes:

[0027] S51: Coordinates (x, y, z) of the current simulation point P; project the current simulation point P onto the neighboring section S to obtain a projection point; based on the coordinates of the projection point, calculate the horizontal search range of the neighboring section as (x±t) or (y±t);

[0028] S52: Based on the search range in the horizontal direction, calculate the distribution depth d of the attribute value of the projection point P0 on the neighboring section;

[0029] S53: Determine a precise search area based on the distribution depth d and the search range (x±t) or (y±t) in the horizontal direction;

[0030] S54: Using the precise search area as a training image for point P;

[0031] S55: Scan the training image using the multi-point statistical template T to obtain a probability distribution function of the training image.

[0032] Optionally, step S5 further includes:

[0033] If the current simulation point P has a neighboring profile S in only one adjacent direction, then the two adjacent complete profiles are recorded as N0 and N1; it is known that there are m attribute values in total on these two profiles;

[0034] Project point P onto neighboring sections N0 and N1 to obtain projection coordinates N0 (d0, z0) and N1 (d1, z1); record the attribute values m0 and m1 of the projection coordinates on N0 and N1;

[0035] Traverse all nodes on the neighbor profile N0 and determine the upper and lower bounds of the intersection of the distribution of attributes m0 and m1, which are N 0sup , N 0inf ; Traverse all nodes on the neighbor profile N1 and determine the upper and lower bounds of the intersection of the distribution of attributes m0 and m1, which are N 1_sup , N 1_inf ; Then for the current simulation point P, the search range established in depth is:

[0036]

[0037] For the projection coordinates N0(d0, z0) and N1(d1, z1), the training image is constrained by the horizontal search range to determine the precise search area, and two parallel training images are obtained as follows:

[0038]

[0039] where d i Represents the horizontal coordinate value of the projection point of point P on the i-th section;

[0040] If the current simulation point P has orthogonal neighbor profiles in two adjacent directions, then two exploration profiles parallel to the projection vector direction are added, and each node to be simulated has four adjacent profiles N0, N1, M0, and M1;

[0041] Project point P onto four adjacent sections N0, N1, M0, and M1 to obtain four projection coordinates;

[0042] Set a horizontal search range value t, and obtain the horizontal search range of the neighbor profile as (x±t) or (y±t);

[0043] Based on the horizontal search range, calculate the distribution depth d of the attribute value of the projection point P0 on the neighboring section;

[0044] According to the distribution depth d and the search range (x±t) or (y±t) in the horizontal direction, the precise search area is determined to obtain an orthogonal training image.

[0045] This application adopts the above technical solution. In the adaptive search strategy, a horizontal search range is prioritized to highlight spatial correlation, and the optimal vertical search range (precise search area) is obtained through spatial projection. The intersection of the horizontal and vertical search ranges is used as a training image for the current node, and multiple adaptive training images are obtained according to this strategy.

[0046] Two types of aggregation methods are employed for different training images: an additive aggregation operator is used for parallel sections to expand the number of training samples, while a multiplicative aggregation operator is used for orthogonal sections to preserve anisotropic features. During the aggregation, the area of the training image and the distance from the section to the simulated point are used as weighting factors to reflect the control of spatial correlation on the simulation. Through probability distribution aggregation of the training images, simulated attribute values of the simulated points are obtained by randomly selecting values from the aggregation function according to probability. An adaptive search strategy leverages the control of stratigraphic structures and other structures in the depth domain to quickly identify optimal local spatial training images, significantly accelerating simulation while improving the reliability of pattern extraction. A probabilistic fusion mechanism comprehensively considers the effects of anisotropy, spatial correlation, and data scale, using spatial distance and image area as key factors in the aggregation formula, thus improving the method's consistency with geological knowledge. Adaptive search and training image aggregation enable the efficient and accurate reproduction of multiple types of three-dimensional spatially heterogeneous patterns in a non-stationary background, manifesting itself at the macro level as the efficient and automated construction of a three-dimensional structure-attribute model.

[0047] Optionally, step S6 includes:

[0048] Combine two parallel training images using the additive aggregation operator:

[0049]

[0050] Among them, P G (A) represents the probability distribution after aggregation by the additive aggregation operator; Pi(A) represents the probability distribution of the i-th training image;

[0051] Among them, A represents the attribute category of probability fusion, and the aggregation weight of two parallel training images is expressed as follows:

[0052]

[0053] In the weight formula, s1 and s2 represent the area of a single training image; d1 and d2 represent the distance from the training image to the current simulation point P;

[0054] Use the multiplication aggregation operator to combine orthogonal training images;

[0055] Assume that the prior probability distribution of the current simulation point P is P0(A), and the fusion weight of the current simulation point P is set to w, where w represents the weight of the probability distribution of n groups of orthogonal directions. Then the joint probability distribution is written as:

[0056]

[0057] Among them, P G (A) represents the probability distribution after aggregation by the additive aggregation operator; P o (A) λ Indicates that the prior probability distribution is fused using λ as the weight; Indicates that w is used as the weight for the i-th training image; where the values of λ and w must not exceed 1 and not be less than 0;

[0058] The aggregate weights of training images in two orthogonal directions are expressed as follows:

[0059]

[0060] In the weight formula, s1 and s2 represent the area of a single training image; d1 and d2 represent the distance from the training image to the current simulation point P.

[0061] An electronic device includes a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs a multi-point statistical three-dimensional stratum modeling method based on adaptive search.

[0062] A computer-readable storage medium stores instructions. When the instructions are executed, a multi-point statistical three-dimensional stratum modeling method based on adaptive search is executed.

[0063] The beneficial effects of the technical solution provided by this application are:

[0064] 1. The present invention uses two-dimensional profiles as basic data instead of the three-dimensional training images required by conventional multi-point statistical methods, thereby achieving efficient automatic reconstruction of complex three-dimensional geological structure models.

[0065] 2. An adaptive search strategy fully embodies the principle of spatial correlation. An adjustable horizontal parameter initially constrains the search range, while an adaptive vertical search domain ensures that contact relationships are correctly captured. Together, these strategies select more accurate and reliable training images. The boundaries of these training images contain macrostructural features and microheterogeneous patterns that are more easily captured than in the initial geological profiles, thus eliminating the need for the multi-point statistical simulation process to rely on the assumption of geological stationarity.

[0066] 3. A probabilistic aggregation mechanism is used to fuse the prior probability distribution, the parallel training image probability distribution, and the orthogonal training image probability distribution. Based on the area and distance of the training images, each distribution maintains a reasonable weight during the aggregation process. Importance is inversely proportional to distance, and the number of samples is proportional to area. The fusion mechanism that associates anisotropy with direction takes into account spatial correlation, anisotropy, and sample size. The joint probability distribution prevents local features from being lost or deviated. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The present application will be further described below with reference to the accompanying drawings and embodiments, in which:

[0068] Figure 1 is a step diagram in an embodiment of the present application;

[0069] Figure 2 This is a schematic diagram of the adaptive search strategy in an embodiment of the present application;

[0070] Figure 3 This is a basic cross-sectional data diagram of an urban geological modeling case in an embodiment of the present application;

[0071] Figure 4 This is the urban geological modeling result diagram based on the DS algorithm in the embodiment of the present application;

[0072] Figure 5 This is a diagram of the urban geological modeling results based on the method of the present invention in an embodiment of the present application;

[0073] Figure 6 is a statistical characteristic diagram of the output results in the embodiment of the present application;

[0074] Figure 7 is a schematic diagram of the structure of an electronic device in an embodiment of the present application;

[0075] Figure 8 is a basic cross-sectional data diagram in the embodiment of the present application;

[0076] Figure 9 It is a three-dimensional stratum model diagram in the embodiment of this application. DETAILED DESCRIPTION

[0077] In order to have a clearer understanding of the technical features, purposes and effects of this application, the specific implementation methods of this application are now described in detail with reference to the accompanying drawings.

[0078] The embodiments of the present application provide a multi-point statistical three-dimensional stratum modeling method based on adaptive search.

[0079] Please refer to Figure 1 , Figure 1 This is a step diagram of a multi-point statistical three-dimensional stratum modeling method based on adaptive search in an embodiment of the present application, including:

[0080] S1: Acquire two-dimensional geological data and assign data points of the two-dimensional geological data to a simulation grid;

[0081] S2: Set the parameters of the simulation grid and the simulation path;

[0082] S3: Traverse the simulation path in a preset order to obtain the current simulation point P; if it is determined that the traversal is not completed, jump to step S4; if it is determined that the traversal is completed, jump to step S8;

[0083] S4: Obtain the coordinates of the current simulation point P and the neighbor profile closest to the current simulation point P;

[0084] S5: traverse the neighbor profile to obtain the probability distribution function of the training image;

[0085] S6: Perform probability aggregation on the probability distribution function to obtain a joint probability distribution function;

[0086] S7: Determine the simulation attribute value of the current simulation point P through the joint probability distribution function; jump to step S3;

[0087] In the specific implementation of the present application, a random value is taken from the joint probability distribution function and used as the simulated attribute value of the current simulation point P.

[0088] S8: Save the current simulation results and complete the three-dimensional reconstruction of all unknown positions on the simulation grid.

[0089] Step S1 includes:

[0090] The two-dimensional geological data includes: horizontal and vertical geological profile data, remote sensing images and drilling sampling data.

[0091] Step S2 includes:

[0092] The parameters of the simulation grid include: the size of the simulation grid and the size of the multi-point statistical template T;

[0093] The simulation path covers all unknown nodes on the current simulation grid.

[0094] Step S4 includes:

[0095] The nearest section includes: a neighbor section closest to the point P in the X-axis direction or a neighbor section closest to the point P in the Y-axis direction.

[0096] Step S5 includes:

[0097] S51: Coordinates (x, y, z) of the current simulation point P; project the current simulation point P onto the neighboring section S to obtain a projection point; based on the coordinates of the projection point, calculate the horizontal search range of the neighboring section as (x±t) or (y±t);

[0098] S52: Based on the search range in the horizontal direction, calculate the distribution depth d of the attribute value of the projection point P0 on the neighboring section;

[0099] In this application, each projected point in geological space has attributes, such as density, lithology, etc.

[0100] S53: Determine a precise search area based on the distribution depth d and the search range (x±t) or (y±t) in the horizontal direction;

[0101] S54: Using the precise search area as a training image for point P;

[0102] S55: Scan the training image using the multi-point statistical template T to obtain a probability distribution function of the training image.

[0103] Step S5 further includes:

[0104] As an embodiment, the determination of the training image based on the adaptive search is described in further detail herein. There are two cases for the flexible and adaptive selection of the training image for a current simulation node P.

[0105] If the current simulation point P has a neighboring profile S in only one adjacent direction, then the two adjacent complete profiles are recorded as N0 and N1; it is known that there are m attribute values in total on these two profiles;

[0106] Project point P onto neighboring sections N0 and N1 to obtain projection coordinates N0 (d0, z0) and N1 (d1, z1); record the attribute values m0 and m1 of the projection coordinates on N0 and N1;

[0107] In a specific implementation of the present application, a property may take different values, for example, density may be 1.2, 2, 3.3, etc. The m property values here refer to the property range, and m0 and m1 are two specific property values.

[0108] Traverse all nodes on the neighbor profile N0 and determine the upper and lower bounds of the intersection of the distribution of attributes m0 and m1, which are N 0sup , N 0inf ; Traverse all nodes on the neighbor profile N1 and determine the upper and lower bounds of the intersection of the distribution of attributes m0 and m1, which are N 1_sup , N 1_inf ; Then for the current simulation point P, the search range established in depth is:

[0109]

[0110] As an example, if the search range is not restricted in the horizontal direction parallel to plane N0, heterogeneous patterns in remote areas may still be captured, but these remote patterns are unreliable. Therefore, a horizontal search range value t is set to impose stricter constraints on the training images.

[0111] For the projection coordinates N0(d0, z0) and N1(d1, z1), the training image is constrained by the horizontal search range to determine the precise search area, and two parallel training images are obtained as follows:

[0112]

[0113] where d i Represents the horizontal coordinate value of the projection point of point P on the i-th section;

[0114] As an embodiment, as shown in the attached Figure 2 The figure illustrates how to gradually determine the adaptive search area as a training image for the two neighboring profiles of the current simulation point P.

[0115] If the current simulation point P has orthogonal neighbor profiles in two adjacent directions, two exploration profiles parallel to the projection vector direction are added, and the space is divided into 9 sub-areas;

[0116] Each node to be simulated has four adjacent sections N0, N1, M0, and M1;

[0117] Project point P onto four adjacent sections N0, N1, M0, and M1 to obtain four projection coordinates;

[0118] Set a horizontal search range value t, and obtain the horizontal search range of the neighbor profile as (x±t) or (y±t);

[0119] Based on the horizontal search range, calculate the distribution depth d of the attribute value of the projection point P0 on the neighboring section;

[0120] According to the distribution depth d and the search range (x±t) or (y±t) in the horizontal direction, the precise search area is determined to obtain an orthogonal training image.

[0121] As an example, according to the above-described adaptive search strategy, the range of the training image is strictly constrained. To preserve contact features, the boundary scanning rule is more gentle. When scanning the boundary of the training image, if the boundary is not empty, the data template is allowed to partially exceed the boundary to capture the pattern, thereby obtaining sufficient information.

[0122] Step S6 includes:

[0123] As an embodiment, the probability fusion strategy is described as follows. Due to the adaptive search, the sizes of multiple sets of training images of each point to be simulated are not exactly the same, and the distance from the node to be simulated to each training image is also different. Therefore, when multiple probabilities are fused, the area and distance of the training image are used as two control factors.

[0124] Combine two parallel training images using the additive aggregation operator:

[0125]

[0126] Among them, P G (A) represents the probability distribution after aggregation by the additive aggregation operator; Pi(A) represents the probability distribution of the i-th training image;

[0127] Among them, A represents the attribute category of probability fusion, and the aggregation weight of two parallel training images is expressed as follows:

[0128]

[0129] In the weight formula, s1 and s2 represent the area of a single training image; d1 and d2 represent the distance from the training image to the current simulation point P;

[0130] Use the multiplication aggregation operator to combine orthogonal training images;

[0131] As an example, anisotropy is widely present in underground structures, so a multiplication aggregation operator is used to fuse orthogonal training images to preserve anisotropic features.

[0132] Assume that the prior probability distribution of the current simulation point P is P0(A), and the fusion weight of the current simulation point P is set to w, where w represents the weight of the probability distribution of n groups of orthogonal directions. Then the joint probability distribution is written as:

[0133]

[0134] Among them, P G (A) represents the probability distribution after aggregation by the additive aggregation operator; P o (A) λ Indicates that the prior probability distribution is fused using λ as the weight; Indicates that w is used as the weight for the i-th training image; where the values of λ and w must not exceed 1 and not be less than 0;

[0135] The aggregate weights of training images in two orthogonal directions are expressed as follows:

[0136]

[0137] In the weight formula, s1 and s2 represent the area of a single training image; d1 and d2 represent the distance from the training image to the current simulation point P.

[0138] In a specific implementation of this application, Figure 3 、 4 5 is an experimental case implemented by the present invention to illustrate the superiority of its method. Figure 3 The four sections used in the experiment are shown (two sections in each of the X and Y directions, forming a bounding box structure). Figure 4 and Figure 5 Two reconstruction results of two different methods (DS method and the method of the present invention) are shown respectively.

[0139] In order to highlight the comparative analysis of the statistical characteristics of the modeling results, each method performed 20 reconstructions in this experimental case, and the statistical characteristics of the 20 output results are as follows: Figure 6 As shown in Figure 2. Spatial connectivity is an important criterion for measuring the quality of reconstruction of 3D geological structure models. Figure 6 (a) shows the spatial connectivity curves of the reconstruction results of the two methods in the X, Y, and Z directions. From the figure, we can see that the connectivity of the output model of the method of the present invention in all directions is closer to the given three-dimensional reference model. Another measure of geological structure anisotropy is the variation of the attribute. Figure 6 (b) is the overall variogram curve of the reconstruction results of the two methods, and the results also show that the method of the present invention has a stronger advantage in the ability to reconstruct the heterogeneity of geological structures. Figure 6 (c) is a comparison curve of the computational efficiency of the two methods as the grid scale increases. It can be seen that the acceleration ratio of the computational efficiency of the method of the present invention gradually increases as the number of simulation grids increases. Therefore, this method is capable of implementing the automatic reconstruction of large-scale fine geological structure models.

[0140] In a specific embodiment of the present application, using Figure 8 The five geological sections shown (three sections in the E direction and two sections in the N direction) were modeled and tested based on the method of the present invention. The size of the model is 60*200*100. Figure 9 The figure shows the 3D modeling results, which also show the distribution of strata of different ages. As can be seen from the figure, this method has a good ability to reconstruct strata containing faults, folds, and pinch-out structures.

[0141] This application also discloses an electronic device. Figure 7 , Figure 7Schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0142] The communication bus 502 is used to implement the connection and communication between these components.

[0143] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.

[0144] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0145] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned multi-point statistical three-dimensional stratum modeling method based on adaptive search.

[0146] The above are merely exemplary embodiments of the present disclosure and are not intended to limit the scope of the present disclosure. In other words, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure.

[0147] This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not described herein. The description and examples are to be considered as exemplary only, and the scope and spirit of the present disclosure are to be defined by the claims.

Claims

1. A multi-point statistical three-dimensional stratum modeling method based on adaptive search, characterized in that: The method comprises the following steps: S1: Acquire two-dimensional geological data and assign data points of the two-dimensional geological data to a simulation grid; S2: Set the parameters of the simulation grid and the simulation path; S3: Traverse the simulation path in a preset order to obtain the current simulation point P; if it is determined that the traversal is not completed, jump to step S4; if it is determined that the traversal is completed, jump to step S8; S4: Obtain the coordinates of the current simulation point P and the neighbor profile closest to the current simulation point P; S5: traverse the neighbor profile to obtain the probability distribution function of the training image; S6: Perform probability aggregation on the probability distribution function to obtain a joint probability distribution function; S7: Determine the simulation attribute value of the current simulation point P through the joint probability distribution function; jump to step S3; S8: Save the current simulation results and complete the three-dimensional reconstruction of all unknown positions on the simulation grid.

2. The method for multi-point statistical three-dimensional stratum modeling based on adaptive search according to claim 1, characterized in that: Step S1 includes: The two-dimensional geological data includes: horizontal and vertical geological profile data, remote sensing images and drilling sampling data.

3. The method for multi-point statistical three-dimensional stratum modeling based on adaptive search according to claim 1, characterized in that: Step S2 includes: The parameters of the simulation grid include: the size of the simulation grid and the size of the multi-point statistical template T; The simulation path covers all unknown nodes on the current simulation grid.

4. The method for multi-point statistical three-dimensional stratum modeling based on adaptive search according to claim 1, characterized in that: Step S4 includes: The nearest section includes: a neighbor section closest to the point P in the X-axis direction or a neighbor section closest to the point P in the Y-axis direction.

5. The method for multi-point statistical three-dimensional stratum modeling based on adaptive search according to claim 3, characterized in that: Step S5 includes: S51: Coordinates (x, y, z) of the current simulation point P; project the current simulation point P onto the neighboring section S to obtain a projection point; based on the coordinates of the projection point, calculate the horizontal search range of the neighboring section as (x±t) or (y±t); S52: Based on the search range in the horizontal direction, calculate the distribution depth d of the attribute value of the projection point P0 on the neighboring section; S53: Determine a precise search area based on the distribution depth d and the search range (x±t) or (y±t) in the horizontal direction; S54: Using the precise search area as a training image for point P; S55: Scan the training image using the multi-point statistical template T to obtain a probability distribution function of the training image.

6. The method for multi-point statistical three-dimensional stratum modeling based on adaptive search according to claim 5, characterized in that: Step S5 further includes: If the current simulation point P has a neighboring profile S in only one adjacent direction, then the two adjacent complete profiles are recorded as N0 and N1; it is known that there are m attribute values in total on these two profiles; Project point P onto neighboring sections N0, N1 to obtain the projection coordinates N 0(d0,z0) ,N 1(d1,z1) ;Record the attribute values m0 and m1 of the projection coordinates on N0, N1; Traverse all nodes on the neighbor profile N0 and determine the upper and lower bounds of the intersection of the distribution of attributes m0 and m1, which are N 0_sup ,N 0_inf ; Traverse all nodes on the neighbor profile N1 and determine the upper and lower bounds of the intersection of the distribution of attributes m0 and m1, which are N 1_sup ,N 1_inf ; Then for the current simulation point P, the search range established in depth is: Projection coordinate N 0(d0,z0) ,N 1(d1,z1) , through the horizontal search range, the training image is constrained and the precise search area is determined. The two parallel training images are expressed as: where d i Represents the horizontal coordinate value of the projection point of point P on the i-th section; If the current simulation point P has orthogonal neighbor profiles in two adjacent directions, then two exploration profiles parallel to the projection vector direction are added, so each node to be simulated has four adjacent profiles N0, N1, M0, M1; Project point P onto four adjacent sections N0, N1, M0, M1 to obtain four projection coordinates; Set a horizontal search range value t, and obtain the horizontal search range of the neighbor profile as (x±t) or (y±t); Based on the horizontal search range, calculate the distribution depth d of the attribute value of the projection point P0 on the neighboring section; According to the distribution depth d and the search range (x±t) or (y±t) in the horizontal direction, the precise search area is determined to obtain an orthogonal training image.

7. The method for multi-point statistical three-dimensional stratum modeling based on adaptive search according to claim 1, characterized in that: Step S6 includes: Combine two parallel training images using the additive aggregation operator: Among them, P G (A) represents the probability distribution after aggregation by the additive aggregation operator; P i (A) represents the probability distribution of the i-th training image; Among them, A represents the attribute category of probability fusion, and the aggregation weight of two parallel training images is expressed as follows: In the weight formula, s1 and s2 represent the area of a single training image; d1 and d2 represent the distance from the training image to the current simulation point P; Use the multiplication aggregation operator to combine orthogonal training images; Assume that the prior probability distribution of the current simulation point P is P0(A), and the fusion weight of the current simulation point P is set to w, where w represents the weight of the probability distribution of n groups of orthogonal directions. Then the joint probability distribution is written as: Among them, P G (A) represents the probability distribution after aggregation by the additive aggregation operator; P o (A) λ Indicates that the prior probability distribution is fused using λ as the weight; Indicates that w is used as the weight for the i-th training image; where the values of λ and w must not exceed 1 and not be less than 0; The aggregate weights of training images in two orthogonal directions are expressed as follows: In the weight formula, s1 and s2 represent the area of a single training image; d1 and d2 represent the distance from the training image to the current simulation point P.

8. An electronic device, characterized in that: The electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 7 is executed.

Images