Multi-point statistical three-dimensional ore body automatic modeling method based on feature enhancement strategy

Abstract

The present invention provides a multi-point statistical three-dimensional ore body automatic modeling method based on a feature enhancement strategy. The method proposes three key strategies in different links of the modeling process to enhance the reasonable reproduction of spatial heterogeneous patterns. The stability enhancement strategy projects all mine profile data onto three planes according to three orthogonal directions of the Cartesian coordinate system, obtains the minimum ellipse surrounding the ore body feature on the plane and expands it to three elliptical cylinders, and the intersection of the elliptical cylinders is used as the area for subsequent simulation. The automatic optimization strategy extracts the ore body distribution characteristics from the initial reconstruction model and only adds it as conditional data to the subsequent simulation. The iterative mechanism makes the new round of ore body reconstruction model gradually close to the attribute distribution state of the original profile until it is optimal. The artifact elimination strategy uses the elimination template for post-processing, and counts the proportion of the ore body within the template range of each simulated node, thereby determining whether the node is a random artifact and retaining or updating it.

Description

Technical Field

[0001] The invention belongs to the field of three-dimensional ore body modeling, and mainly relates to a multi-point statistical three-dimensional ore body automatic modeling method based on a feature enhancement strategy. Background Art

[0002] In the field of geological resource exploration, 3D modeling and visualization technology plays an increasingly important role. For deep-buried solid mineral resources, ore body models are constructed using exploration data to express the spatial distribution of ore bodies, reveal the distribution patterns of ore bodies, and infer the mineralization geological background, which is helpful for studying the formation mechanism of ore deposits and ultimately providing useful strategic guidance for various production work in mines.

[0003] As one of the theories of three-dimensional geological modeling, multi-point geostatistics has become an important branch in the field of ore body modeling. Various multi-point geostatistics specific algorithms have been widely proposed, showing good prospects. As a high-order statistical method, multi-point geostatistics first proposed to bring spatial heterogeneous patterns from training images into the target model. The data template is a specific template composed of multiple grids in multiple vector directions around a single center point. The training image is a type of concrete image containing specific geological heterogeneous patterns. Through the data template, the patterns and probability distributions in the prior training images are reproduced in the target model. For most multi-point geostatistical algorithms, such as the SNESIM algorithm, the training image is scanned through the data template, and the heterogeneous patterns are stored in the search tree or search list. Inspired by the text pattern simulation in the information field, the DS algorithm directly captures similar or identical heterogeneous patterns from the training image to achieve reconstruction, and the pattern distance is used to measure the degree of similarity between the heterogeneous patterns in the training image and the data events.

[0004] Although there are many existing multi-point geostatistical methods, given the complexity of the evolution and formation of ore bodies in mines, prior training images, as an important prerequisite for multi-point geostatistical methods, are very difficult to obtain in the actual reconstruction of three-dimensional ore body models. Compared with complete three-dimensional training images, data such as exploration lines, boreholes and geological profiles are widely available in the exploration work for mines and are easier to obtain. In order to overcome the strong dependence on three-dimensional training images, some multi-point geostatistical algorithms that reconstruct three-dimensional models using two-dimensional data have been proposed. For example, the multi-point statistical three-dimensional geological structure automatic reconstruction method based on local search strategy divides the simulation grid into several subdomains through multiple profiles in different orthogonal directions, and the corresponding sub-profiles in each subdomain are used as training images. Through the aggregation probability, simulation is achieved.

[0005] The multi-point statistical algorithm for reconstructing a three-dimensional geological model from a two-dimensional profile has solved the problem of scarce training images to a certain extent. However, this type of method is still difficult to be extended to the actual ore body modeling application in mines. On the one hand, although the geological data such as boreholes, profiles, and exploration lines obtained in the exploration work are abundant, they are still very sparse relative to the three-dimensional geological space of the entire mining area. The limited exploration data means that spatial heterogeneous patterns, such as ore body distribution patterns, are difficult to be fully and reliably captured, counted, and reproduced. On the other hand, multi-point geological statistical methods usually assume the stationarity assumption by default, that is, it is believed that the distribution of heterogeneous patterns in training images and target models is uniform. However, due to the complexity of its generation and evolution process, real ore bodies often show strong non-stationarity. The aforementioned multi-point statistical algorithms often find it difficult to overcome non-stationarity in the actual ore body modeling process, thereby simulating a large number of artifacts, which cannot match the heterogeneous pattern distribution of conditional data, and it is difficult to reproduce the spatial characteristics of training images. It is difficult to make a reasonable geological interpretation of the modeling results, and it is even more difficult to guide actual production work.

[0006] In view of the shortcomings and challenges faced by the existing multi-point geological statistical stochastic modeling methods in actual ore body modeling applications, it is necessary to propose a multi-point statistical three-dimensional ore body automatic modeling method based on feature enhancement strategy, so that in the actual three-dimensional geological modeling work, the geological non-stationarity problem faced by model reconstruction can be overcome as much as possible, and at the same time, the multi-category statistical information contained in the profile data can be fully utilized to automatically generate the optimal ore body model to meet the needs of actual mining exploration and development work. Summary of the invention

[0007] In order to solve the shortcomings of the existing process of reconstructing three-dimensional geological models based on two-dimensional profiles, the present invention provides a multi-point statistical three-dimensional ore body automatic modeling method based on feature enhancement. The method proposes three key strategies in different links of ore body construction from the three aspects of non-stationarity, spatial continuity and spatial correlation to enhance the reasonable reproduction of spatial heterogeneous patterns.

[0008] The present invention provides a multi-point statistical three-dimensional ore body automatic modeling method based on a feature enhancement strategy, the method comprising the following steps:

[0009] S1: Load the mining area exploration profile data and distribute the known data to the simulation grid;

[0010] S2: Project the known data onto the XOY, YOZ, and XOZ planes. For the ore body grade characteristics on the plane, find out the convex hull C1, C2, and C3. For C1, C2, and C3, calculate their minimum circumscribed ellipse R1, R2, and R3. Extend the minimum circumscribed ellipse to three dimensions to obtain the circumscribed elliptical cylinder V1, V2, and V3, and find their intersection V as the subsequent ore body modeling space.

[0011] S3: Calculate the prior probability set P0 of all sub-domains divided by the horizontal and vertical sections according to the position of each section data in the simulation grid;

[0012] S4: If it is the first round of simulation, go directly to step S5, otherwise load the ore grade dataset D in the previous round of model M Loaded into the 3D grid as known condition data;

[0013] S5: define a random simulation path sim_path for V, so that it covers all nodes to be simulated in V. Obtain the nodes to be simulated x from the simulation path sim_path in sequence, and determine the corresponding data events in three orthogonal directions according to the specific position of x in the three-dimensional simulation grid;

[0014] S6: Obtain all n sub-sections around x as training images, and scan the sub-sections with the three data events in step S5 in turn to obtain corresponding n conditional probability distribution functions cpdfs;

[0015] S7: Obtain the prior probability p0 of the subdomain corresponding to x from P0, fuse the above n conditional probability distribution functions cpdfs and the prior probability p0 of the subdomain corresponding to x, and obtain the joint probability distribution;

[0016] S8: Randomly select a value from the joint probability distribution and assign it to the current simulation position x as its simulation attribute;

[0017] S9: If the traversal is not completed, go to step S6 to simulate and obtain values ​​for other nodes to be simulated in sim_path, otherwise go to step S10;

[0018] S10: After simulating all unknown nodes, a three-dimensional model M is obtained; an elimination template E is defined, and a path E_path is set to sequentially traverse all nodes in V;

[0019] S11: Get the node t to be processed, record its attribute Pt, traverse and eliminate the nodes in the template E, and count the attribute types and the proportion of each attribute;

[0020] S12: if the attribute Pt of t has the highest proportion in the template, the attribute of t is retained, otherwise its attribute is updated to the attribute with the highest proportion in the range of the three-dimensional elimination template;

[0021] S13: If the traversal is not completed, go to S11, otherwise go to S14;

[0022] S14: for a specific attribute, calculate its distribution ratio Pro on the conditional section; for the post-processed three-dimensional model M0, calculate its attribute distribution ratio Prm on the reconstructed section;

[0023] S15: Given a similarity evaluation threshold s, if |Pro - Prm| < s, output the optimized model M0; otherwise, extract specific attributes from M0 to form a dataset D M , and go to step S4.

[0024] The beneficial effects of the present invention are as follows:

[0025] (1) The stability enhancement strategy provided by the present invention focuses on the macroscopic distribution characteristics of the ore body during the modeling process, divides and demarcates the entire mining area to be simulated regionally, thereby greatly enhancing the stability of the local space; at the same time, since the surrounding rock area outside the ore body is distinguished, the overall modeling efficiency is greatly improved.

[0026] (2) The artifact elimination strategy adopted in the present invention uses an elimination template to largely distinguish and process the artifacts generated during the stochastic modeling process. The shape and size of the elimination template can be adjusted, which means that it can be flexibly adjusted according to the actual situation in different types of ore body modeling application scenarios, and more reasonably process the artifacts.

[0027] (3) The adaptive optimization strategy in the present invention counts the distribution ratio of the ore body in the conditional data and uses it as the main parameter for automatically seeking optimization during the iterative process of the reconstruction model, making full and reasonable use of the information contained in the low-dimensional data itself. The direct effect is that the reconstructed optimal ore body model is closest to the original data of the mining area in terms of the distribution ratio of the ore body.

[0028] (4) The application of the automatic iteration mechanism in the present invention adds the simulated micro-ore body as soft data to the iterative simulation, which solves the problem of scarce exploration data to a certain extent and makes the present invention more robust. Description of the Drawings

[0029] Figure 1 is the overall process of the present invention.

[0030] Figure 2 is a schematic diagram of the working process of the present invention in an actual ore body modeling application. Figure 2 (a) is the data preprocessing process; Figure 2 (b) is the process of ore body modeling using the present invention; Figure 2 (c) is the analysis and application of the ore body model.

[0031] Figure 3 is a schematic diagram of the process of obtaining the minimum circumscribed ellipse of specific attributes of the stability enhancement strategy in the present invention on a two-dimensional plane. Figure 3 (a) is the projection process; Figure 3 (b) is the process of determining the convex hull on the ore body projection plane; Figure 3 (c) is the process of calculating the minimum circumscribed circle of the convex hull; Figure 3(d) is the process of calculating the minimum circumscribed ellipse of the convex hull.

[0032] Figure 4 It is a schematic diagram of the process of delineating the ore body distribution area in a three-dimensional grid using the stability enhancement strategy in the present invention. Figure 4 (a) is the projection process; Figure 4 (b) and (c) are the projection processes;

[0033] Figure 5 It is a schematic diagram of the process of post-processing the ore body model using the elimination template in the artifact elimination strategy of the present invention. Figure 5 (a) are different elimination templates; Figure 5 (b) is the model to be processed after artifact removal; Figure 5 (c) is the process of statistical update by eliminating template judgment; Figure 5 (d) is the ore body model after artifacts are eliminated.

[0034] Figure 6 It is a schematic diagram of the model adaptive optimization strategy in the present invention using an iterative mechanism to construct an optimal ore body model.

[0035] Figure 7 It is an ore body reconstruction model generated by the present invention using exploration data of a certain mining area.

[0036] Figure 8 The invention uses the method to reconstruct 20 ore body models, extracts three original condition sections and four adjacent reconstructed sections to draw variogram curves, wherein the variograms of the reconstructed sections are collectively expressed as confidence intervals.

[0037] Fig. 9 It is a multi-dimensional scaling analysis diagram for 20 ore body models reconstructed by the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0039] Before explaining the present invention in a better way, the key points of the present invention are described again as follows:

[0040] In actual ore body modeling applications, strong non-stationarity is often encountered, which is contrary to the stationarity assumption of the usual multi-point statistical modeling method. Therefore, after all the exploration profile data are placed in the simulation grid, the simulation area is first delineated using the stationarity enhancement strategy. That is, all profile data are projected onto three planes in three orthogonal directions of the Cartesian coordinate system, and various attributes on these three planes are superimposed. Select one or some features on the plane and find the minimum ellipse that surrounds the feature on the plane. For the ellipses on the three planes, three elliptical cylinders are obtained by expanding to three dimensions. The intersection of the elliptical cylinders is the area for subsequent simulation. In this area, the stationarity of the selected features is significantly improved. Reconstructing a three-dimensional model from a two-dimensional profile means the scarcity of data, which leads to very limited reproduction of heterogeneous patterns in a single simulation in practical applications.

[0041] An adaptive optimization strategy based on an iterative mechanism is used to obtain a reconstruction model that best matches the original exploration profile data of the mining area. For the initial simulation implementation, the main attribute features are extracted from it and only added as conditional data to the subsequent simulation process to ensure that the pattern extraction in the subsequent simulation is not affected. As the iterative mechanism proceeds, the reconstructed profile in the reconstruction model will gradually approach the attribute distribution state of the original geological profile. After each iteration, the ratio of the ore body on the reconstructed profile to the original profile is calculated. When the difference between the two ratios is less than a predetermined threshold, it means that the reconstructed profile and the original profile have achieved a high degree of consistency in ore body distribution, and the reconstruction model is the optimal ore body model.

[0042] The multi-point statistical method is random, which manifests itself as randomly generated artifacts in the specific simulation implementation. The characteristics of this type of artifact include isolated existence and random distribution, which does not conform to the principle of spatial continuity. Similar to the data template in multi-point statistics, an elimination template is used in the artifact elimination strategy for post-processing of the model. The elimination template consists of a central node and several nodes around the central node in different vector directions. It is not used to count patterns, but to count various attributes within the spatial range contained in the template. In the post-processing process, for each simulated node, the proportion of each attribute in the elimination template is statistically calculated. If the attribute of the central node is not the attribute with the largest proportion in the template, it can be considered that the node belongs to a random artifact, and its attribute is updated to the attribute with the largest proportion.

[0043] The above three key strategies are aimed at enhancing the selected features in the simulation process, so as to achieve better application results in the actual ore body modeling work and provide strong methodological guarantee and technical support. In ore body modeling work, such selected features generally refer to ore bodies.

[0044] The present invention provides a multi-point statistical three-dimensional ore body automatic modeling method based on feature enhancement strategy. The method proposes three key strategies in different links of the modeling process to enhance the reasonable reproduction of spatial heterogeneous patterns. The stability enhancement strategy projects all profile data onto three planes according to three orthogonal directions of the Cartesian coordinate system, obtains the minimum ellipse surrounding certain features on the plane and expands it to three elliptical cylinders, and the intersection of the elliptical cylinders is used as the area for subsequent simulation. The automatic optimization strategy extracts the main attribute features from the initial reconstruction model and only adds them as conditional data to the subsequent simulation. The iterative mechanism makes the new round of reconstruction model gradually close to the ore body distribution state of the original profile until it is optimal. The artifact elimination strategy uses the elimination template for post-processing, and counts the proportion of each attribute of each simulated node within the template range, thereby determining whether the node is a random artifact and retaining or updating it.

[0045] Please refer to Figure 1-Figure 2 , the method includes the following process in an actual ore body modeling work in southwest China:

[0046] S1: Load geological profile data and distribute known data to simulation grids; in the present invention, the grid size is set to 200×200×200;

[0047] S2: Figure 3 As shown, the known data are projected onto the XOY, YOZ, and XOZ planes. For the ore body grade characteristics on the plane, the convex hulls C1, C2, and C3 are found. For C1, C2, and C3, their minimum circumscribed ellipses R1, R2, and R3 are calculated. The minimum circumscribed ellipse is extended to three dimensions to obtain the circumscribed elliptical cylinders V1, V2, and V3, and their intersection V is calculated as the subsequent ore body modeling space, as shown in FIG. Figure 4 As shown;

[0048] S3: Calculate the prior probability set P0 of all sub-domains divided by the horizontal and vertical sections according to the position of each section data in the simulation grid;

[0049] S4: If it is the first round of simulation reconstruction, go directly to step S5, otherwise load the ore grade dataset D in the previous round of model M Loaded into the 3D grid as known condition data;

[0050] S5: define a random simulation path sim_path for V, so that it covers all nodes to be simulated in V. Obtain the nodes to be simulated x from the simulation path sim_path in sequence, and determine the corresponding data events in three orthogonal directions according to the specific position of x in the three-dimensional simulation grid;

[0051] S6: Obtain all n sub-profiles around x as training images, and sequentially scan the sub-profiles with the three data events in step S5 to obtain the corresponding n conditional probability distribution functions cpdfs;

[0052] S7: Obtain the prior probability p0 of the sub-domain corresponding to x from P0, fuse the above n conditional probability distribution functions cpdfs and the prior probability p0 of the sub-domain corresponding to x, and obtain the joint probability distribution;

[0053] S8: Randomly take values from the joint probability distribution and assign them to the current simulation position x as its simulation attribute;

[0054] S9: If the traversal is not complete, go to step S6 to perform simulation value taking on other nodes to be simulated in sim_path, otherwise go to step S10;

[0055] S10: After all unknown nodes are simulated, obtain a reconstructed three-dimensional model M; define an elimination template E, and set a path E_path to sequentially traverse all nodes in V;

[0056] S11: As Figure 5 shown, obtain the node t to be processed, record its attribute Pt, traverse the nodes in the elimination template E, and count the types of attributes and the proportion of each attribute;

[0057] S12: If the proportion of the attribute Pt of t in the template is the highest, retain the attribute of t, otherwise update its attribute to the attribute with the highest proportion within the range of the three-dimensional elimination template;

[0058] S13: If the traversal is not complete, go to S11, otherwise go to S14;

[0059] S14: For a specific attribute, calculate its distribution proportion Pro on the conditional profile; for the reconstructed model M0 after post-processing, calculate the attribute distribution proportion Prm on its reconstructed profile;

[0060] S15: Given the similarity evaluation threshold s, if |Pro - Prm| < s, output the optimized model M0, otherwise extract the specific attributes in M0 to form the data set D M , and go to step S4.

[0061] As Figure 6 shown, in the reconstruction of this ore body, with the increase of the number of iterations and the application of the artifact elimination strategy, the shape and contour of the ore body gradually approach the training image and match the actual ore body distribution. As Figure 7 shown, after the reconstructed model of this ore body is placed in the actual geological environment through transformation, it fits the ground surface accurately, the shape of the ore body is clear, the distribution is concentrated, and the spatial distribution of the ore body is faithfully reproduced.

[0062] In order to analyze the statistical characteristics of the modeling results, 20 reconstructions were performed using the method of the present invention in this modeling example, and the 20 output results were plotted as multidimensional scaling analysis diagrams, such as Fig. 9 As shown in the figure, as a random modeling method, the present invention has good stability in multiple simulations. Further, for each model, three conditional profiles and their adjacent 12 reconstructed profiles are proposed and subjected to variance analysis. The variance function of the reconstructed profile is integrated as Figure 8 It can be seen that the variogram curve of the conditional section is distributed in the center of the confidence band, and the extension direction of the confidence band is consistent with the variogram curve of the conditional section. This shows that the heterogeneous pattern and statistical characteristics of the conditional data are well reproduced in the reconstruction model, and the generated model can also support actual geological modeling work.

[0063] The beneficial effects of the present invention are:

[0064] (1) The stability enhancement strategy provided by the present invention focuses on the macroscopic distribution characteristics of the ore body during the reconstruction process, and regionally divides and delineates the entire mining area to be simulated, thereby greatly enhancing the stability of the local space; at the same time, since the surrounding rock area outside the ore body is distinguished, the overall modeling efficiency is greatly improved.

[0065] (2) The artifact elimination strategy adopted in the present invention uses an elimination template to distinguish and process the artifacts generated by the random modeling process to a large extent. The shape and size of the elimination template can be adjusted, which means that it can be flexibly adjusted according to actual conditions in different types of ore body modeling application scenarios, and artifacts can be processed more reasonably.

[0066] (3) The adaptive optimization strategy in the present invention counts the distribution ratio of the ore bodies in the conditional data and uses it as the main parameter for automatically seeking the optimal implementation in the iterative process of reconstructing the model, making full and reasonable use of the information contained in the low-dimensional data itself. The direct result is that the reconstructed optimal ore body model is closest to the original data of the mining area in terms of the distribution ratio of the ore bodies.

[0067] (4) The use of the automatic iteration mechanism in the present invention adds the simulated micro-ore body as soft data to the iterative simulation, which solves the problem of scarce exploration data to a certain extent and makes the present invention more robust.

Claims

1. A multi-point statistical three-dimensional ore body automatic modeling method based on feature enhancement strategy, characterized by: The following steps are involved: S1: Load the mining area exploration profile data and distribute the known data to the simulation grid; S2: Project the known data onto the XOY, YOZ, and XOZ planes. For the ore body grade characteristics on the plane, find out the convex hull C1, C2, and C3. For C1, C2, and C3, calculate their minimum circumscribed ellipse R1, R2, and R3. Extend the minimum circumscribed ellipse to three dimensions to obtain the circumscribed elliptical cylinder V1, V2, and V3, and find their intersection V as the subsequent ore body modeling space. S3: Calculate the prior probability set P0 of all sub-domains divided by the horizontal and vertical sections according to the position of each section data in the simulation grid; S4: If it is the first round of simulation, go directly to step S5, otherwise load the ore grade dataset D in the previous round of model M Loaded into the 3D grid as known condition data; S5: define a random simulation path sim_path for V, so that it covers all nodes to be simulated in V; obtain the nodes to be simulated x from the simulation path sim_path in sequence, and determine the corresponding data events in three orthogonal directions according to the specific position of x in the three-dimensional simulation grid; S6: Obtain all n sub-sections around x as training images, and scan the sub-sections with the three data events in step S5 in turn to obtain corresponding n conditional probability distribution functions cpdfs; S7: Obtain the prior probability p0 of the subdomain corresponding to x from P0, fuse the above n conditional probability distribution functions cpdfs and the prior probability p0 of the subdomain corresponding to x, and obtain the joint probability distribution; S8: Randomly select a value from the joint probability distribution and assign it to the current simulation position x as its simulation attribute; S9: If the traversal is not completed, go to step S6 to simulate and obtain values ​​for other nodes to be simulated in sim_path, otherwise go to step S10; S10: After all unknown nodes are simulated, a reconstructed three-dimensional ore body model M is obtained; an elimination template E is defined, and a path E_path that sequentially traverses all nodes in V is set; S11: Get the node t to be processed, record its attribute Pt, traverse and eliminate the nodes in the template E, and count the attribute types and the proportion of each attribute; S12: if the attribute Pt of t has the highest proportion in the template, the attribute of t is retained, otherwise its attribute is updated to the attribute with the highest proportion in the range of the three-dimensional elimination template; S13: If the traversal is not completed, go to S11, otherwise go to S14; S14: For a specific attribute, calculate its distribution ratio Pro on the conditional section; for the post-processed ore body model M0, calculate its attribute distribution ratio Prm on the reconstructed section; S15: Given a similarity evaluation threshold s, if |Pro - Prm| < s, output the optimized model M0; otherwise, extract specific attributes from M0 to form a data set D M , and go to step S4.

Images