A three-dimensional multi-parameter fusion igneous rock positioning method and system

Abstract

This invention discloses a three-dimensional multi-parameter fusion method and system for igneous rock location, relating to the field of geological exploration. The method includes: acquiring three-dimensional geological parameter data of a target area and preprocessing it to obtain three-dimensional multi-parameter data; using a trained igneous rock classification and recognition model to identify the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data; and performing smoothing processing based on the probability prediction values ​​of the three-dimensional grid types and the probability prediction values ​​of neighboring three-dimensional grid types to determine the final type of all three-dimensional grids, thereby obtaining the igneous rock mass type and three-dimensional location. This invention, by aligning and fusing multiple three-dimensional data of the target area and confirming the rock mass type of each three-dimensional grid based on multiple parameters, can accurately identify the rock mass type at each location, thus improving the accuracy of igneous rock mass location.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, and more specifically to a three-dimensional multi-parameter fusion method and system for locating magmatic rocks. Background Technology

[0002] Igneous rocks are formed when magma erupts onto the Earth's surface or intrudes into the Earth's crust, cools, and solidifies. Common igneous rocks include granite, syenite, diorite, andesite, gabbro, and basalt. Locating igneous rocks is of great significance for geological structure research, mineral exploration, and geological dating.

[0003] Existing methods for locating igneous rocks typically involve directly inverting the location of underground rock masses using electrical and electromagnetic methods, or inverting the distribution of underground rock strata through seismic exploration, or using gravity measuring instruments to depict the gravity field distribution and thus determine the approximate location of the igneous rock mass. However, relying on a single method to locate igneous rock masses suffers from insufficient positioning accuracy and poor precision.

[0004] Therefore, how to improve the accuracy of igneous rock mass positioning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a three-dimensional multi-parameter fusion method and system for igneous rock positioning, which improves the positioning accuracy of igneous rock masses by acquiring multiple three-dimensional parameters of the target area.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention discloses a three-dimensional multi-parameter fusion method for locating igneous rocks, the specific steps of which are as follows:

[0008] Acquire three-dimensional data of various geological parameters in the target area and preprocess them to obtain three-dimensional multi-parameter data;

[0009] Using a trained igneous rock classification and recognition model, the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data is determined;

[0010] Based on the probability prediction values ​​of the type of the 3D mesh and the probability prediction values ​​of the types of neighboring 3D meshes, smoothing is performed to determine the final type of all 3D meshes, thus obtaining the igneous rock mass type and 3D location.

[0011] Furthermore, the three-dimensional geological parameter data includes: three-dimensional density data, three-dimensional magnetic susceptibility data, three-dimensional resistivity data, three-dimensional polarizability data, three-dimensional elastic modulus data, three-dimensional mineral alteration data, and three-dimensional radioactive gamma spectroscopy measurement data, which are obtained through the following methods:

[0012] Divide the target area into multiple sub-regions;

[0013] Gravity field data of each sub-region is obtained using a gravity measuring instrument, and underground three-dimensional density data is inverted.

[0014] The geomagnetic field of each of the sub-regions was measured using a magnetometer, and the three-dimensional magnetic susceptibility data of the subsurface was retrieved.

[0015] The three-dimensional resistivity and three-dimensional polarizability data of each sub-region were measured and retrieved using electrical and electromagnetic methods.

[0016] Seismic wave data of each sub-region was obtained through seismic exploration, and the three-dimensional elastic modulus data of the subsurface was inverted.

[0017] Acquire satellite hyperspectral remote sensing data for each of the sub-regions and invert three-dimensional mineral alteration data;

[0018] Radioactive exploration was conducted on each of the sub-regions, and underground three-dimensional radioactive gamma spectrum measurement data were retrieved.

[0019] Furthermore, dividing the target area into multiple sub-regions includes: measuring radon and mercury in the target area to determine the location and orientation of deep fractures in the target area; and dividing the target area into multiple sub-regions with independent structures based on the location and orientation of the deep fractures.

[0020] Furthermore, the preprocessing yields three-dimensional multi-parameter data, including:

[0021] The three-dimensional data of the geological parameters are cleaned to remove erroneous data and missing data are interpolated and supplemented.

[0022] Set the resampling 3D grid size and resample the cleaned 3D geological parameter data;

[0023] The three-dimensional density data, the three-dimensional magnetic susceptibility data, the three-dimensional resistivity data, the three-dimensional polarizability data, the three-dimensional elastic modulus data, the three-dimensional mineral alteration data, and the three-dimensional radioactive gamma spectroscopy measurement data are resampled and aligned with each three-dimensional grid, and then fused to obtain the three-dimensional multi-parameter data.

[0024] Furthermore, the igneous rock classification and recognition model is obtained by stacking multiple restricted Boltzmann machines, and the specific training method is as follows:

[0025] The multi-parameter data of the actual material corresponding to the type is determined by experiments. The parameter values ​​of each parameter are used as sample features, and the material type is used as the sample label to obtain a training sample set.

[0026] The restricted Boltzmann machines are pre-trained using the training sample set.

[0027] Multiple restricted Boltzmann machines that have been pre-trained are superimposed to form an initial igneous rock classification and recognition model;

[0028] The igneous rock classification and identification model is fine-tuned using the backpropagation algorithm to update the network parameters of the initial igneous rock classification and identification model, thus obtaining the final identification model.

[0029] Furthermore, the pre-training of multiple restricted Boltzmann machines is performed using the contrastive divergence algorithm, specifically as follows:

[0030] The training sample set is input into the restricted Boltzmann machine to determine the state of the visible units of the training sample set in the restricted Boltzmann machine and the probability of the hidden units in the restricted Boltzmann machine.

[0031] Based on the states and probabilities, a first joint activation probability and a second joint activation probability are calculated; the first joint activation probability is used to characterize the joint activation probability of visible units and hidden units under the data distribution; the second joint activation probability is used to characterize the joint activation probability of the reconstructed distribution.

[0032] The weights of the restricted Boltzmann machine from the visible units to the hidden units are updated based on the first joint activation probability and the second joint activation probability.

[0033] Furthermore, when updating the weights from visible units to hidden units, the incremental formula for weight updates is:

[0034] Δw qxy =∈ ra ( <v x h y > data - <v x h y > recon );

[0035] Where, Δw qxy For the weight update increment of the restricted Boltzmann machine, w qxy Represents the weights from the x-th visible unit to the y-th hidden unit in the q-th layer; ∈ ra The learning rate; data The symbol representing the joint activation probability of visible and hidden units under the data distribution; v x h represents the state of the x-th visible unit. y Represents a given visible state v x The probability of the y-th hidden unit at time t; reconThe symbol represents the joint activation probability under the reconstructed distribution.

[0036] Furthermore, the backpropagation algorithm is a backpropagation algorithm based on a loss function, wherein the loss function is:

[0037]

[0038] Among them, C k Let N be the loss function for the igneous rock classification and identification model, where N is the total number of samples and K is the total number of categories; t nk y represents the true label of the k-th target category of the n-th sample; nk λ represents the predicted probability that the nth sample belongs to the kth class; bh w is the regularization coefficient of the loss function. qxy This represents the weights of the q-th layer of the recognition model from the x-th visible unit to the y-th hidden unit. The square of the weight.

[0039] Furthermore, the smoothing process specifically includes:

[0040] Step 1: Sort the probability prediction values ​​of each 3D mesh according to its type;

[0041] Step 2: Select a 3D grid to be determined, and determine whether the difference between the largest and second largest values ​​after sorting is greater than a set threshold;

[0042] Step 3: If the value is greater than the predicted probability, the type corresponding to the maximum value of the predicted probability will be used as the final type of the 3D mesh; otherwise, proceed to Step 4.

[0043] Step 4: Calculate the probability prediction mean and variance of each type of the neighboring three-dimensional meshes of the three-dimensional mesh to be determined, and determine the probability correction value of the three-dimensional mesh to be determined;

[0044] Step 5: Sort all probability correction values ​​of the three-dimensional mesh to be determined, and select the type corresponding to the maximum value as the final type of the three-dimensional mesh;

[0045] Step 6: Determine whether the final type of all 3D meshes has been determined. If yes, end the smoothing process; otherwise, return to step 2.

[0046] The formula for the probability correction value is:

[0047]

[0048] in, P represents the probability correction value for the j-th type of the 3D mesh to be determined; oj This represents the probability prediction value of the j-th type of the three-dimensional mesh to be determined; The probability prediction average of the j-th type of neighboring 3D meshes is represented; e is a natural constant; The probability prediction variance of the j-th type of neighboring 3D mesh is expressed by the formula:

[0049]

[0050] Among them, P ij Let represent the probability prediction value of the j-th type of the i-th neighboring 3D mesh, where i = 1, 2, 3...M, and M is the total number of neighboring 3D meshes.

[0051] This invention also discloses a three-dimensional multi-parameter fusion igneous rock positioning system, comprising:

[0052] Data acquisition module: Acquires three-dimensional data of various geological parameters in the target area and preprocesses them to obtain three-dimensional multi-parameter data;

[0053] Model recognition module: Using a trained igneous rock classification and recognition model, determine the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data;

[0054] Rock mass positioning module: Based on the probability prediction values ​​of the type of the 3D mesh and the probability prediction values ​​of the types of neighboring 3D meshes, smoothing is performed to determine the final type of all 3D meshes, thus obtaining the igneous rock mass type and 3D positioning.

[0055] As can be seen from the above technical solution, compared with the prior art, this invention discloses a three-dimensional multi-parameter fusion method and system for igneous rock positioning. By performing fracture measurement and regional division of the target area, it can adapt to complex geological environments, improving its flexibility and effectiveness in practical applications. By fusing multiple geological parameters, it can more comprehensively reflect the geological characteristics of the target area, helping to eliminate the limitations of single methods, improve the reliability of overall analysis, and thus significantly improve the positioning accuracy of igneous rock masses. Furthermore, using a trained igneous rock classification and recognition model, it can not only locate igneous rock masses but also classify and identify different igneous rocks. This invention, by aligning and fusing multiple three-dimensional data of the target area and confirming the rock mass type of each three-dimensional grid based on multiple parameters, can accurately confirm the rock mass type at each location, thereby improving the positioning accuracy of igneous rock masses. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention. Detailed Implementation

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] This invention discloses a three-dimensional multi-parameter fusion method for locating magmatic rocks, such as... Figure 1 As shown, the specific steps are as follows:

[0060] Acquire three-dimensional data of various geological parameters in the target area and preprocess them to obtain three-dimensional multi-parameter data;

[0061] Using a trained igneous rock classification and identification model, the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data is determined;

[0062] Based on the probability prediction values ​​of the type of the 3D mesh and the probability prediction values ​​of the types of neighboring 3D meshes, smoothing is performed to determine the final type of all 3D meshes, thus obtaining the igneous rock mass type and 3D location.

[0063] In one specific embodiment, the three-dimensional geological parameter data includes: three-dimensional density data, three-dimensional magnetic susceptibility data, three-dimensional resistivity data, three-dimensional polarizability data, three-dimensional elastic modulus data, three-dimensional mineral alteration data, and three-dimensional radioactive gamma spectroscopy measurement data, which are obtained through the following methods:

[0064] Divide the target area into multiple sub-regions;

[0065] Gravity field data for each sub-region was obtained using gravity measuring instruments, and underground three-dimensional density data was inverted.

[0066] The geomagnetic field of each sub-region was measured using a magnetometer, and the three-dimensional magnetic susceptibility data of the subsurface was retrieved.

[0067] The three-dimensional resistivity and three-dimensional polarizability data of each sub-region were measured and retrieved using electrical and electromagnetic methods.

[0068] Seismic wave data of each sub-region was obtained through seismic exploration, and the three-dimensional elastic modulus data of the subsurface was inverted.

[0069] Acquire satellite hyperspectral remote sensing data for each sub-region and invert three-dimensional mineral alteration data;

[0070] Radioactive exploration was conducted in each sub-region, and the underground three-dimensional radioactive gamma spectrum measurement data were retrieved. The above-mentioned inversion methods based on gravity field, geomagnetic field, electrical method, electromagnetic method, and seismic wave are commonly used methods in the field of geological exploration, and will not be elaborated further.

[0071] In one specific embodiment, the target area is divided into multiple sub-regions, including: measuring radon and mercury in the target area to determine the location and orientation of deep faults; and dividing the target area into multiple sub-regions with independent structures based on the location and orientation of the deep faults. Radon and mercury often release to the surface along deep fault zones. By measuring the distribution and concentration of these gases, the location and orientation of deep faults can be delineated. By dividing the area into multiple sub-regions, the geological characteristics of each region can be analyzed in more detail, and specific geological anomalies or features, especially those related to magmatic activity, can be better identified. This reduces the interference of faults and fissures on the inversion process and improves the accuracy of the inverted three-dimensional density, magnetic susceptibility, resistivity, polarizability, and elastic modulus data.

[0072] In one specific embodiment, preprocessing yields three-dimensional multi-parameter data, including:

[0073] Data cleaning is performed on the three-dimensional geological parameter data to remove erroneous data and interpolate and supplement missing data;

[0074] Set the resampling 3D grid size and resample the cleaned 3D geological parameter data;

[0075] The three-dimensional grids in the resampled three-dimensional density data, three-dimensional magnetic susceptibility data, three-dimensional resistivity data, three-dimensional polarizability data, three-dimensional elastic modulus data, three-dimensional mineral alteration data, and three-dimensional radioactive gamma spectroscopy measurement data are aligned and registered, and then fused to obtain three-dimensional multi-parameter data.

[0076] In one specific embodiment, the igneous rock classification and recognition model is obtained by stacking multiple restricted Boltzmann machines, and the specific training method is as follows:

[0077] The multi-parameter data of the actual materials corresponding to the type were determined by experiments. The parameter values ​​of each parameter were used as sample features, and the material type was used as the sample label to obtain a training sample set.

[0078] Multiple restricted Boltzmann machines were pre-trained using the training sample set;

[0079] Multiple restricted Boltzmann machines that have been pre-trained are superimposed to form an initial igneous rock classification and recognition model;

[0080] The backpropagation algorithm was used to fine-tune the igneous rock classification and recognition model, update the network parameters of the initial igneous rock classification and recognition model, and obtain the final recognition model.

[0081] In one specific embodiment, the types of the three-dimensional mesh include: granite, syenite, diorite, andesite, gabbro, basalt, as well as air, water, and soil.

[0082] In one specific embodiment, the contrastive divergence algorithm is used to pre-train multiple restricted Boltzmann machines, specifically as follows:

[0083] The training sample set is input into the Restricted Boltzmann Machine (RBM) to determine the state of the training sample set in the visible units of the RBM and the probability of it in the hidden units of the RBM.

[0084] Based on the state and probability, the first joint activation probability and the second joint activation probability are calculated; the first joint activation probability is used to characterize the joint activation probability of visible and hidden units under the data distribution; the second joint activation probability is used to characterize the joint activation probability of the reconstructed distribution.

[0085] The weights from visible units to hidden units of the restricted Boltzmann machine are updated based on the first joint activation probability and the second joint activation probability.

[0086] In a specific embodiment, when updating the weights from visible units to hidden units, the incremental formula for weight updates is:

[0087] Δw qxy =∈ ra ( <v x h y > data - <v x h y > recon );

[0088] Where, Δw qxy For the weight update increment of the restricted Boltzmann machine, w qxy Represents the weights from the x-th visible unit to the y-th hidden unit in the q-th layer; ∈ ra The learning rate; data The symbol representing the joint activation probability of visible and hidden units under the data distribution; v x h represents the state of the x-th visible unit. y Represents a given visible state v x The probability of the y-th hidden unit at time t; recon The symbol represents the joint activation probability under the reconstructed distribution.

[0089] The formula for the joint activation probability of visible and hidden units under the data distribution is:

[0090]

[0091] Where N is the total number of samples, v nx v represents the state of the x-th visible unit in the n-th sample. nx It is the state of the xth visible unit; h ny (v nx ) represents a given visible state v nx When the probability of the nth sample to the yth hidden unit is given by the formula:

[0092]

[0093] Where Sig represents the sigmoid activation function, w xy b represents the weight from the x-th visible unit to the y-th hidden unit. y The bias of the y-th hidden unit.

[0094] The formula for the joint activation probability of the reconstructed distribution is:

[0095]

[0096] in, This represents the state of the x-th visible unit in the n-th sample obtained through hidden state reconstruction; Represents a given hidden state When, the probability of the nth sample to the yth hidden unit.

[0097] In a specific embodiment, the backpropagation algorithm is a loss function-based backpropagation algorithm, where the loss function is:

[0098]

[0099] Among them, C k Let N be the loss function for the igneous rock classification and identification model, where N is the total number of samples and K is the total number of categories; t nk y represents the true label of the k-th target category of the n-th sample; nk λ represents the predicted probability that the nth sample belongs to the kth class; bh w is the regularization coefficient of the loss function. qxy This represents the weights of the q-th layer of the recognition model from the x-th visible unit to the y-th hidden unit. The square of the weight.

[0100] In a specific embodiment, the smoothing process is as follows:

[0101] Step 1: Sort the probability prediction values ​​of each 3D mesh type;

[0102] Step 2: Select a 3D grid to be determined, and determine whether the difference between the largest and second largest values ​​after sorting is greater than a set threshold;

[0103] Step 3: If the value is greater than the predicted value, the type corresponding to the maximum value of the probability prediction is taken as the final type of the 3D mesh; otherwise, proceed to Step 4.

[0104] Step 4: Calculate the probability prediction mean and variance of each type of the neighboring 3D meshes to be determined, and determine the probability correction value of the 3D mesh to be determined.

[0105] Step 5: Sort all probability correction values ​​of the 3D mesh to be determined, and select the type corresponding to the maximum value as the final type of the 3D mesh;

[0106] Step 6: Determine whether the final type of all 3D meshes has been determined. If yes, end the smoothing process; otherwise, return to step 2.

[0107] The formula for the probability correction value is:

[0108]

[0109] in, P represents the probability correction value for the j-th type of the 3D mesh to be determined; oj This represents the probability prediction value of the j-th type of the 3D mesh to be determined; The probability prediction average of the j-th type in the neighboring 3D grid is represented; e is the natural constant. The variance of the probability prediction for the j-th type of the neighboring 3D mesh is expressed by the formula:

[0110]

[0111] Among them, P ij Let represent the probability prediction value of the j-th type of the i-th neighboring 3D grid, where i = 1, 2, 3…M, and M is the total number of neighboring 3D grids.

[0112] By applying threshold judgments and corrections to the predicted probabilities of various types of igneous rocks output by the igneous rock classification and identification model, the probability of each grid can be corrected and smoothed based on the neighboring three-dimensional grids, thereby further improving the accuracy of identification and location of various types of igneous rocks.

[0113] This invention also discloses a three-dimensional multi-parameter fusion magmatic rock positioning system, comprising:

[0114] Data acquisition module: Acquires three-dimensional data of various geological parameters in the target area and preprocesses them to obtain three-dimensional multi-parameter data;

[0115] Model recognition module: Using a trained igneous rock classification and recognition model, determine the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data;

[0116] Rock mass positioning module: Based on the probability prediction values ​​of the type of the 3D mesh and the probability prediction values ​​of the types of neighboring 3D meshes, smoothing is performed to determine the final type of all 3D meshes, thus obtaining the igneous rock mass type and 3D positioning.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0118] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A three-dimensional multi-parameter fusion method for locating igneous rocks, characterized in that, The specific steps are as follows: Acquire three-dimensional data of various geological parameters in the target area and preprocess them to obtain three-dimensional multi-parameter data; Using a trained igneous rock classification and recognition model, the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data is determined; Based on the probability prediction values ​​of the types of the 3D meshes and the probability prediction values ​​of the types of neighboring 3D meshes, a smoothing process is performed to determine the final type of all 3D meshes, thus obtaining the igneous rock mass type and 3D location; the smoothing process specifically includes: Step 1: Sort the probability prediction values ​​of each 3D mesh according to its type; Step 2: Select a 3D grid to be determined, and determine whether the difference between the largest and second largest values ​​after sorting is greater than a set threshold; Step 3: If the value is greater than the predicted probability, the type corresponding to the maximum value of the predicted probability will be used as the final type of the 3D mesh; otherwise, proceed to Step 4. Step 4: Calculate the probability prediction mean and variance of each type of the neighboring three-dimensional meshes of the three-dimensional mesh to be determined, and determine the probability correction value of the three-dimensional mesh to be determined; Step 5: Sort all probability correction values ​​of the three-dimensional mesh to be determined, and select the type corresponding to the maximum value as the final type of the three-dimensional mesh; Step 6: Determine whether the final type of all 3D meshes has been determined. If yes, end the smoothing process; otherwise, return to step 2. The formula for the probability correction value is: in, P represents the probability correction value for the j-th type of the 3D mesh to be determined; oj This represents the probability prediction value of the j-th type of the three-dimensional mesh to be determined; The probability prediction average of the j-th type of neighboring 3D meshes is represented; e is a natural constant; The probability prediction variance of the j-th type of neighboring 3D mesh is expressed by the formula: Among them, P ij Let represent the probability prediction value of the j-th type of the i-th neighboring 3D mesh, where i = 1, 2, 3...M, and M is the total number of neighboring 3D meshes.

2. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 1, characterized in that, The three-dimensional geological parameter data includes: three-dimensional density data, three-dimensional magnetic susceptibility data, three-dimensional resistivity data, three-dimensional polarizability data, three-dimensional elastic modulus data, three-dimensional mineral alteration data, and three-dimensional radioactive gamma spectroscopy measurement data, which are obtained through the following methods: Divide the target area into multiple sub-regions; Gravity field data of each sub-region is obtained using a gravity measuring instrument, and underground three-dimensional density data is inverted. The geomagnetic field of each of the sub-regions was measured using a magnetometer, and the three-dimensional magnetic susceptibility data of the subsurface was retrieved. The three-dimensional resistivity and three-dimensional polarizability data of each sub-region were measured and retrieved using electrical and electromagnetic methods. Seismic wave data of each sub-region was obtained through seismic exploration, and the three-dimensional elastic modulus data of the subsurface was inverted. Acquire satellite hyperspectral remote sensing data for each of the sub-regions and invert three-dimensional mineral alteration data; Radioactive exploration was conducted on each of the sub-regions, and underground three-dimensional radioactive gamma spectrum measurement data were retrieved.

3. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 2, characterized in that, The step of dividing the target area into multiple sub-regions includes: measuring radon and mercury in the target area to determine the location and orientation of deep fractures in the target area; and dividing the target area into multiple sub-regions with independent structures based on the location and orientation of the deep fractures.

4. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 2, characterized in that, The preprocessing yields three-dimensional multi-parameter data, including: The three-dimensional data of the geological parameters are cleaned to remove erroneous data and missing data are interpolated and supplemented. Set the resampling 3D grid size and resample the cleaned 3D geological parameter data; The three-dimensional density data, the three-dimensional magnetic susceptibility data, the three-dimensional resistivity data, the three-dimensional polarizability data, the three-dimensional elastic modulus data, the three-dimensional mineral alteration data, and the three-dimensional radioactive gamma spectroscopy measurement data are resampled and aligned with each three-dimensional grid, and then fused to obtain the three-dimensional multi-parameter data.

5. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 1, characterized in that, The igneous rock classification and recognition model is obtained by stacking multiple restricted Boltzmann machines. The specific training method is as follows: The multi-parameter data of the actual material corresponding to the type is determined by experiments. The parameter values ​​of each parameter are used as sample features, and the material type is used as the sample label to obtain a training sample set. The restricted Boltzmann machines are pre-trained using the training sample set. Multiple restricted Boltzmann machines that have been pre-trained are superimposed to form an initial igneous rock classification and recognition model; The igneous rock classification and identification model is fine-tuned using the backpropagation algorithm to update the network parameters of the initial igneous rock classification and identification model, thus obtaining the final identification model.

6. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 5, characterized in that, The pre-training of multiple restricted Boltzmann machines is performed using the contrastive divergence algorithm, specifically as follows: The training sample set is input into the restricted Boltzmann machine to determine the state of the visible units of the training sample set in the restricted Boltzmann machine and the probability of the hidden units in the restricted Boltzmann machine. Based on the states and probabilities, a first joint activation probability and a second joint activation probability are calculated; the first joint activation probability is used to characterize the joint activation probability of visible units and hidden units under the data distribution; the second joint activation probability is used to characterize the joint activation probability of the reconstructed distribution. The weights of the restricted Boltzmann machine from the visible units to the hidden units are updated based on the first joint activation probability and the second joint activation probability.

7. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 6, characterized in that, When updating the weights from visible units to hidden units, the incremental formula for weight updates is: Δw qxy =∈ ra (<v x h y > data -<v x h y > recon ); Where, Δw qxy For the weight update increment of the restricted Boltzmann machine, w qxy Represents the weights from the x-th visible unit to the y-th hidden unit in the q-th layer; ∈ ra The learning rate; data The symbol representing the joint activation probability of visible and hidden units under the data distribution; v x h represents the state of the x-th visible unit. y Represents a given visible state v x The probability of the y-th hidden unit at time t; recon The symbol represents the joint activation probability under the reconstructed distribution.

8. The three-dimensional multi-parameter fusion method for igneous rock localization according to claim 5, characterized in that, The backpropagation algorithm is a backpropagation algorithm based on a loss function, which is: Among them, C k Let N be the loss function for the igneous rock classification and identification model, where N is the total number of samples and K is the total number of categories; t nk y represents the true label of the k-th target category of the n-th sample; nk λ represents the predicted probability that the nth sample belongs to the kth class; bh w is the regularization coefficient of the loss function. qxy This represents the weights of the q-th layer of the recognition model from the x-th visible unit to the y-th hidden unit. The square of the weight.

9. A three-dimensional multi-parameter fusion igneous rock positioning system, characterized in that, include: Data acquisition module: Acquires three-dimensional data of various geological parameters in the target area and preprocesses them to obtain three-dimensional multi-parameter data; Model recognition module: Using a trained igneous rock classification and recognition model, determine the probability prediction value of the type of each three-dimensional grid in the three-dimensional multi-parameter data; Rock mass location module: Based on the probability prediction values ​​of the types of the 3D meshes and the probability prediction values ​​of the types of neighboring 3D meshes, a smoothing process is performed to determine the final type of all 3D meshes, thus obtaining the igneous rock mass type and 3D location. The smoothing process specifically involves: Step 1: Sort the probability prediction values ​​of each 3D mesh according to its type; Step 2: Select a 3D grid to be determined, and determine whether the difference between the largest and second largest values ​​after sorting is greater than a set threshold; Step 3: If the value is greater than the predicted probability, the type corresponding to the maximum value of the predicted probability will be used as the final type of the 3D mesh; otherwise, proceed to Step 4. Step 4: Calculate the probability prediction mean and variance of each type of the neighboring three-dimensional meshes of the three-dimensional mesh to be determined, and determine the probability correction value of the three-dimensional mesh to be determined; Step 5: Sort all probability correction values ​​of the three-dimensional mesh to be determined, and select the type corresponding to the maximum value as the final type of the three-dimensional mesh; Step 6: Determine whether the final type of all 3D meshes has been determined. If yes, end the smoothing process; otherwise, return to step 2. The formula for the probability correction value is: in, P represents the probability correction value for the j-th type of the 3D mesh to be determined; oj This represents the probability prediction value of the j-th type of the three-dimensional mesh to be determined; The probability prediction average of the j-th type of neighboring 3D meshes is represented; e is a natural constant; The probability prediction variance of the j-th type of neighboring 3D mesh is expressed by the formula: Among them, P ij Let represent the probability prediction value of the j-th type of the i-th neighboring 3D mesh, where i = 1, 2, 3...M, and M is the total number of neighboring 3D meshes.

Images