Sample data set generator for deep learning inversion of mine resistivity
Through the sample dataset generator used for deep learning inversion of resistivity in mines, the resistivity distribution of underground media is simulated and training samples are automatically generated, which solves the complexity and noise sensitivity problems of DC resistivity inversion methods in the prior art. , achieving more efficient and reliable inversion results.
Patent Information
- Application Number
- CN202411431280.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-14
- Publication Date
- 2025-02-11
AI Technical Summary
The existing DC resistivity inversion method is complex in calculation, sensitive to the initial model, and susceptible to noise interference, resulting in low uncertainty and reliability of the inversion result.
It provides a sample dataset generator for deep learning inversion of resistivity in mines. By simulating the resistivity distribution of different underground media, it automatically generates a large number of training samples, supporting the training and optimization of deep learning models. The generator includes a model building module, a data simulation module, a calculation module, a processing module and a data generation module. It can build a dielectric resistivity model, simulate the mine working face structure, calculate theoretical resistivity observations, add noise and interference values, and generate Deep learning training data.
By automatically generating high-quality training samples, the training efficiency and optimization speed of deep learning models can be improved, the accuracy and reliability of inversion results can be enhanced, and the cost and risk of field exploration can be reduced.
Smart Images
Figure CN119415947A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mine data processing, and in particular to a sample data set generator for deep learning inversion of mine resistivity. Background Art
[0002] The DC resistivity method is a commonly used technique in geophysical exploration, which infers geological structure and rock properties by measuring the resistivity difference of underground media. However, the traditional DC resistivity inversion method is complex in calculation, sensitive to the initial model, and easily interfered by noise, resulting in uncertainty and low reliability of the inversion results.
[0003] In order to improve the accuracy and efficiency of DC resistivity inversion, most existing DC resistivity inversion systems use deep learning models to perform inversion, thereby providing reliability and accuracy of the inversion. However, the training of deep learning models requires a large amount of high-quality sample data, while the actual underground resistivity data obtained is limited and often accompanied by noise, which limits its application in geological exploration.
[0004] Based on this, there is an urgent need for a sample dataset generator for deep learning inversion of mine resistivity, which can automatically generate a large number of training samples by simulating the resistivity distribution of different underground media to support the training and optimization of deep learning models. Summary of the invention
[0005] One of the purposes of the present invention is to provide a sample data set generator for deep learning inversion of mine resistivity, which can automatically generate a large number of training samples by simulating the resistivity distribution of different underground media to support the training and optimization of deep learning models.
[0006] In order to achieve the above objectives, a sample dataset generator for deep learning inversion of mine resistivity is provided, including:
[0007] The model building module is used to build several dielectric resistivity models based on the existing geological data and geological theories and the preset model building strategy, and customize the resistivity value corresponding to each dielectric resistivity model;
[0008] It is also used to construct a corresponding mine working face structure model based on the preset mine working face construction conditions on the basis of each medium resistivity model;
[0009] The data simulation module is used to simulate the exploration process of DC resistivity in the corresponding mine working face according to the constructed structure models of each mine working face and based on the preset data acquisition simulation strategy, and set the corresponding basic exploration parameters;
[0010] The calculation module is used to calculate the corresponding theoretical resistivity observation value according to the set basic exploration parameters and the corresponding mine working face structure model;
[0011] A processing module is used to identify the corresponding noise level and determine the corresponding noise value according to the data characteristics corresponding to the actual exploration data, and add the noise value to the theoretical resistivity observation value to form the respective corresponding simulated resistivity observation values;
[0012] The data generation module is used to identify the medium resistivity profile corresponding to each simulated resistivity observation value based on the corresponding medium resistivity model, and package each simulated resistivity observation value and its corresponding medium resistivity profile into a specified format as the corresponding deep learning training data.
[0013] Technical principles and effects of this scheme: In this scheme, firstly, the resistivity models of various media are constructed through the existing geological data and geological theories, and the resistivity values corresponding to each medium resistivity model are customized, so as to realize the construction of multiple medium resistivity models.
[0014] Then, based on each medium resistivity model and based on the preset model building strategy, several medium resistivity models are constructed, and the resistivity value corresponding to each medium resistivity model is customized; by simulating the mine working face structure model under different conditions, the generated data set can cover a wider range of situations, which helps to train the model with better generalization ability.
[0015] Afterwards, according to the constructed structural models of each mine working face and based on the preset data acquisition simulation strategy, the exploration process of DC resistivity in the corresponding mine working face is simulated, and the corresponding basic exploration parameters are set; according to the set basic exploration parameters and the corresponding mine working face structural model, the corresponding theoretical resistivity observation value is calculated; compared with traditional exploration methods, the use of this data generator can efficiently complete a large number of simulation experiments in a laboratory environment, reducing the cost and risk of field exploration.
[0016] According to the data characteristics corresponding to the actual exploration data, the corresponding noise level is identified, and the corresponding noise value is determined, and the noise value is added to the theoretical resistivity observation value to form the corresponding simulated resistivity observation value; through the customized geological model and noise addition mechanism, resistivity observation data that is closer to the actual situation can be generated, thereby improving the quality of training data.
[0017] According to each simulated resistivity observation value, based on the corresponding medium resistivity model, the medium resistivity profile corresponding to the simulated resistivity observation value is identified, and each simulated resistivity observation value and its corresponding medium resistivity profile are packaged into a specified format as the corresponding deep learning training data, realizing the automatic generation of a large amount of training data, reducing the time for manual data collection and annotation, and speeding up the model development and iteration. That is, by simulating the resistivity distribution of different underground media, a large number of training samples can be automatically generated to support the training and optimization of deep learning models.
[0018] Furthermore, the preset model building strategy includes the following steps:
[0019] S10, determining the corresponding model range;
[0020] S11, creating a plurality of model grids according to geological principles and existing geological data, and setting the grid density corresponding to the model grid based on the position information of the corresponding preset electrode in the corresponding model grid, wherein the grid density gradually becomes sparser as the distance between the grid and the preset electrode increases;
[0021] S12. Setting different resistivity values for each model network to form each final dielectric resistivity model.
[0022] Beneficial effects: In this scheme, multiple model grids are created and different resistivity values are set for the model grids to achieve diversified construction of the medium resistivity model, providing diverse data for the subsequent process.
[0023] Furthermore, the preset mine working face construction conditions include:
[0024] S20, in each of the formed medium resistivity models, determining a geometric model to be set in a grid corresponding to each of the medium resistivity models;
[0025] S21, generating corresponding two-dimensional resistivity models in batches according to the determined geometric model, and setting corresponding model parameters and the number of two-dimensional resistivity models to be generated;
[0026] S22, mapping the batch-generated two-dimensional resistivity models onto the medium resistivity model, and setting different resistivities for different parts of the mapped two-dimensional resistivity models to form structural models of the working faces of various mines.
[0027] Beneficial effects: In this scheme, a mine working face structure model is constructed based on the resistivity models of various media, so as to truly and accurately simulate the resistivity differences within the mine working face and the impact of the mine working face structure on current propagation.
[0028] Furthermore, the preset data acquisition simulation strategy includes the following steps:
[0029] S30, according to each formed mine working face structure model, discretize the model area corresponding to the mine working face structure model to generate a corresponding grid;
[0030] S31, setting a corresponding power source and a current characteristic corresponding to the power source, and determining the electrode point position and the current injection point corresponding to the electrode arrangement in the model area based on the characteristics of the model area, and associating the electrode at the current injection point with the power source based on the electrode point position and the current injection point;
[0031] S32. After the electrode at the current injection point is associated with the power supply, the corresponding power supply is set to the emission current mode, and current is supplied to the underground medium through the electrode associated with the power supply to establish a stable current field. The voltage difference between each electrode point is calculated, and based on the corresponding voltage difference, data related to the resistivity of the underground medium is obtained.
[0032] Beneficial effects: In this scheme, the model area corresponding to the mine working face structure model is first discretized, and the complex model is simplified into a form that is easy to calculate, ensuring the accuracy and reliability of the calculation. Then, by setting the current characteristics corresponding to the power supply, it is convenient to perform effective electromagnetic field calculations under different simulation conditions, greatly improving the efficiency of subsequent data calculations.
[0033] At the same time, the electrodes are arranged and the current injection points are set according to the characteristics of the model area, so that the positions of the electrode points and the current injection points can be reasonably set, so as to achieve accurate correlation between the electrodes and the power supply at the current injection points, which can ensure that the subsequent simulation can accurately capture the changes in the voltage difference, further improve the stability of the current transmission, and greatly provide the flexibility and sensitivity of the entire system.
[0034] Through accurate and realistic simulation of the DC resistivity exploration process, the exploration task can be completed more economically and efficiently, and the exploration cost can be reduced. At the same time, reliable data can be provided to provide reliable data support for the calculation of subsequent theoretical resistivity observation values.
[0035] Furthermore, the processing module is also used to quantify the interference factors according to the interference factors, calculate the corresponding interference values, and add them to the corresponding simulated resistivity observation values.
[0036] Beneficial effects: In this scheme, not only the influence of noise level on theoretical resistivity observation value is taken into account, but also the influence of interference factors on theoretical resistivity observation value is further considered, thereby achieving a true and accurate reflection of the data characteristics in the actual exploration process, and greatly improving the accuracy and authenticity of simulated resistivity observation value.
[0037] Furthermore, the data generation module includes:
[0038] A data sorting module is used to identify the labels corresponding to the simulated resistivity observation values according to the formed simulated resistivity observation values and the force resistivity observation values corresponding to the simulated resistivity observation values, wherein the labels include resistivity model parameters, interference factors and noise levels corresponding to the simulated resistivity observation values;
[0039] A profile generating module is used to generate a corresponding dielectric resistivity profile according to each simulated resistivity observation value; the dielectric resistivity profile is a three-dimensional dielectric resistivity profile or a two-dimensional dielectric resistivity profile;
[0040] The sample data construction module is used to take the simulated resistivity observation value as the input feature to construct the feature matrix corresponding to the data set, and take the medium resistivity profile corresponding to the simulated resistivity observation value as the output feature to form the corresponding sample data set;
[0041] The format conversion module is used to convert the formed sample data set into a specified format as the corresponding deep learning training data set.
[0042] Beneficial effects: Through unified data collation standards, the consistency and comparability of data sets are guaranteed, and data quality is improved. The generated resistivity profile provides an intuitive display of geological information, which facilitates a quick understanding of underground structures. The constructed standardized sample data set is directly applicable to deep learning model training, reducing the time and complexity of data preprocessing. Supports multiple data format conversions, allowing data to be flexibly migrated between different analysis tools and platforms. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 This is a logic block diagram of a sample data set generator for deep learning inversion of mine resistivity in Embodiment 1 of the present invention.
[0044] Figure 2 This is a schematic diagram of the model of the mine working face structure model in Example 1 of the present invention.
[0045] Figure 3 Schematic diagram of the arrangement of electrode points in the first embodiment of the present invention.
[0046] Figure 4 This is a schematic diagram of a model containing a single abnormal body in Example 1 of the present invention.
[0047] Figure 5 This is a cross-sectional diagram of the dielectric resistivity in Example 1 of the present invention. DETAILED DESCRIPTION
[0048] The following is further described in detail through specific implementation methods:
[0049] Embodiment 1
[0050] A sample dataset generator for deep learning inversion of mine resistivity, basically as follows Figure 1 As shown, including:
[0051] The model building module is used to build several dielectric resistivity models based on the existing geological data and geological theories and the preset model building strategy, and customize the resistivity value corresponding to each dielectric resistivity model;
[0052] The preset model building strategy includes the following steps:
[0053] S10. Determine the corresponding model range; for example, X=1000m, Y=500m.
[0054] S11. According to geological principles and existing geological data, several model grids are created, and the grid density corresponding to the model grid is set based on the position information of the corresponding preset electrode in the corresponding model grid, and the grid density gradually becomes sparse as the distance between the grid and the preset electrode increases; in this embodiment, since the model is a set of parameters that characterize geological characteristics, the model is a vector. After the grid is created, different physical properties can be assigned to each cell to represent different geological conditions or abnormal bodies under geological conditions. In the actual DC electrical data acquisition process, the potential generated by the current electrode decays with the increase of the distance r, and the decay rate is r. -3 Therefore, smaller cells are needed near the current electrodes to accurately simulate the field, while larger cells can be used far away from the current electrodes because the field changes smoothly in these places. According to this situation, a tree grid needs to be generated (the grid density is high at the electrode and the grid becomes sparse as it moves away from the electrode). This is the difficulty in this simulation method. If there are terrain factors, the changes in the grid need to be adjusted according to the different terrain.
[0055] S12, setting different resistivity values for each model network to form each final dielectric resistivity model. In this embodiment, different conductivity values may also be set. The different resistivity values set include air resistivity, background resistivity, abnormal body resistivity, etc.
[0056] The model building module is also used to build a corresponding mine working face structure model based on the various medium resistivity models and the preset mine working face construction conditions; Figure 2 As shown, the schematic diagram corresponding to the corresponding mine working face structure model.
[0057] The preset mine working face construction conditions include:
[0058] S20. In each of the formed medium resistivity models, determine the geometric model that needs to be set in the grid corresponding to each medium resistivity model; for example, establish a model containing a single abnormal body in a uniform medium, or two layers of uniform medium models contain different abnormal bodies. The schematic diagram of the model containing a single abnormal body is as follows: Figure 4 shown.
[0059] S21, generating corresponding two-dimensional resistivity models in batches according to the determined geometric model, and setting corresponding model parameters and the number of two-dimensional resistivity models to be generated;
[0060] S22, mapping the batch-generated two-dimensional resistivity model to the medium resistivity model, and setting different resistivities for different parts of the mapped two-dimensional resistivity model to form the structural model of each mine working face. That is, according to different settings, different resistivity values are assigned to grid cells of different parts of the two-dimensional resistivity model, the background is assigned a background resistivity value, and the abnormal body is assigned an abnormal body resistivity value.
[0061] The data simulation module is used to simulate the exploration process of DC resistivity in the corresponding mine working face according to the constructed structural models of each mine working face and based on the preset data acquisition simulation strategy, and set the corresponding basic exploration parameters; in this embodiment, the simulation of the exploration process of DC resistivity in the corresponding mine working face includes the steps of electrode arrangement, current injection, voltage measurement, etc.
[0062] The preset data acquisition simulation strategy includes the following steps:
[0063] S30, according to each formed mine working face structure model, discretize the model area corresponding to the mine working face structure model to generate a corresponding grid;
[0064] S31, set the corresponding power supply and the current characteristics corresponding to the power supply, and based on the characteristics of the model area, determine the electrode point position and the current injection point corresponding to the electrode arrangement in the model area, and associate the electrode at the current injection point with the power supply based on the electrode point position and the current injection point; in this embodiment, in order to effectively associate the power supply with the electrode point to ensure the stable transmission of the current and the accurate calculation of the voltage difference, accurately define the current characteristics of the power supply (such as frequency, amplitude, etc.) to ensure that the propagation characteristics of the current in the underground medium are consistent with the actual situation. This step can be adjusted according to different geological conditions. When selecting the electrode position, it is necessary to consider the resistivity distribution and geometric characteristics of the underground medium to maximize the measurement accuracy of the voltage difference. Since different electrode spacing selections will result in different detection depths and ranges, it is necessary to accurately determine the geometric position of the power supply and the receiver (in this simulation, since the geometric model is built on a grid-divided plane, the position of the power supply and the receiver can be accurately located). Accurate setting of the grounding and insulation conditions of the electrode. The electrode grounding condition is selected based on the interface between the air or coal seam and the geological body in the established resistivity model. The insulation condition is to set the air or coal seam resistivity to infinity when setting the resistivity, so as to reasonably set the boundary conditions for current injection. Figure 3 shown.
[0065] S32, after the electrode at the current injection point is associated with the power supply, the corresponding power supply is set to the emission current mode, and the current is supplied to the underground medium through the electrode associated with the power supply to establish a stable current field, and the voltage difference between each electrode point is calculated, and based on the corresponding voltage difference, data related to the resistivity of the underground medium is obtained. This step is to set the boundary conditions for current injection and set the constant value to the current.
[0066] In this embodiment, the relationship between the underground resistivity distribution, the stable current field and the potential at each point is determined by Ohm's law. The Ohm's law equation in the form of current vector in a continuous medium is as follows:
[0067] J=E / ρ
[0068] Among them, ρ is the dielectric resistivity, J is the current density, E is the electric field strength, and the relationship between the potential φ and the field strength is:
[0069] E=-Vφ
[0070] The resistivity detection method often uses a point power supply for power supply. Assuming that the power supply is provided at the spatial position (x0, y0, z0), according to the law of conservation of charge, the relationship between the current density J and the current field I can be obtained as follows:
[0071]
[0072] Among them, ΔV is the unit volume, and δ is the Diktral function. According to the above relationship, the relationship between the dielectric resistivity ρ, the current field I and the potential φ can be obtained as follows:
[0073]
[0074] The above equation is the Poisson equation for the potential distribution generated by a point source. In the actual forward modeling process, by adding boundary conditions, the potential distribution φ can be solved through the medium resistivity distribution ρ.
[0075] In the actual measurement process, the power supply electrodes (A, B) are used to supply power to the geological body while the potential difference ΔU is observed between the measuring electrodes (M, N). MN , and according to the current I and the potential difference ΔU MN Calculate the apparent resistivity ρ s , the formula is as follows:
[0076]
[0077] Where K is the device coefficient. Different observation devices have different arrangements of the four electrodes (A, B, M, N), so K is different. The calculation formula is as follows:
[0078]
[0079] The calculation module is used to calculate the corresponding theoretical resistivity observation value according to the set basic exploration parameters and the corresponding mine working face structure model; in this embodiment, first, the mathematical model of the current field is constructed using the current continuity equation and Ohm's law, and then the current field is solved by numerical methods to calculate the propagation of the current in the medium. Considering that a large amount of data needs to be processed at this time, this requires the efficiency and accuracy of the algorithm to avoid result deviations caused by calculation errors. Therefore, optimizing the numerical simulation process and algorithm design is the key to achieving efficient analysis. In this embodiment, a Python-based algorithm is used to implement the entire calculation process.
[0080] Use efficient data structures: Choose to apply NumPy in Python to improve data storage and access efficiency.
[0081] Compressed storage: Compress data for storage to reduce memory usage, especially when processing large-scale data.
[0082] Multithreading / multiprocessing: Use Python's multiprocessing library to distribute data processing tasks to multiple threads or processes to speed up calculations.
[0083] A processing module is used to identify the corresponding noise level and determine the corresponding noise value according to the data characteristics corresponding to the actual exploration data, and add the noise value to the theoretical resistivity observation value to form the respective corresponding simulated resistivity observation values;
[0084] It is also used to quantify the interference factors according to the interference factors, calculate the corresponding interference values, and add them to the corresponding simulated resistivity observation values. In this embodiment, in order to simulate the uncertainty in the actual exploration data, we define a noise model. This usually includes random noise and systematic errors. Random noise can be simulated using Gaussian distribution.
[0085] Then, noise is added to the theoretical resistivity observation. The specific method is to add the randomly generated noise value to the theoretical value to form data that is closer to the actual observation. For example, if the theoretical value is V and the noise is N, the simulated observation value V ′ , which can be expressed as:
[0086] V ′ =V+N
[0087] Interference factors: In addition to random noise, other interference factors need to be considered, such as poor electrode contact, terrain effects, etc. These factors can be quantified by specific interference models and added to the simulation data.
[0088] The data generation module is used to identify the medium resistivity profile corresponding to each simulated resistivity observation value based on the corresponding medium resistivity model, and package each simulated resistivity observation value and its corresponding medium resistivity profile into a specified format as the corresponding deep learning training data.
[0089] The data generation module comprises:
[0090] A data sorting module is used to identify the labels corresponding to the simulated resistivity observation values according to the formed simulated resistivity observation values and the force resistivity observation values corresponding to the simulated resistivity observation values, wherein the labels include resistivity model parameters, interference factors and noise levels corresponding to the simulated resistivity observation values;
[0091] The profile generation module is used to generate a corresponding dielectric resistivity profile according to each simulated resistivity observation value; the dielectric resistivity profile is a three-dimensional dielectric resistivity profile or a two-dimensional dielectric resistivity profile; in this embodiment, these profiles are visualized using Matplotlib to help intuitively understand the resistivity distribution and its changes. The generated dielectric resistivity profile is as follows: Figure 5 shown.
[0092] The sample data construction module is used to take the simulated resistivity observation value as the input feature to construct the feature matrix corresponding to the data set, and take the medium resistivity profile corresponding to the simulated resistivity observation value as the output feature to form the corresponding sample data set;
[0093] The format conversion module is used to convert the formed sample data set into a specified format as the corresponding deep learning training data set. In this embodiment, the sorted data set is converted into a format suitable for a deep learning framework (such as TensorFlow or PyTorch). This may include saving the data as a CSV file, TFRecord format, or other suitable structure.
[0094] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is described too much here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can know all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A sample data set generator for deep learning inversion of mine resistivity, characterized by: include: The model building module is used to build several dielectric resistivity models based on the existing geological data and geological theories and the preset model building strategy, and customize the resistivity value corresponding to each dielectric resistivity model; It is also used to construct a corresponding mine working face structure model based on the preset mine working face construction conditions on the basis of each medium resistivity model; The data simulation module is used to simulate the exploration process of DC resistivity in the corresponding mine working face according to the constructed structure models of each mine working face and based on the preset data acquisition simulation strategy, and set the corresponding basic exploration parameters; The calculation module is used to calculate the corresponding theoretical resistivity observation value according to the set basic exploration parameters and the corresponding mine working face structure model; A processing module is used to identify the corresponding noise level and determine the corresponding noise value according to the data characteristics corresponding to the actual exploration data, and add the noise value to the theoretical resistivity observation value to form the respective corresponding simulated resistivity observation values; The data generation module is used to identify the medium resistivity profile corresponding to each simulated resistivity observation value based on the corresponding medium resistivity model, and package each simulated resistivity observation value and its corresponding medium resistivity profile into a specified format as the corresponding deep learning training data.
2. A sample data set generator for deep learning inversion of mine resistivity according to claim 1, characterized in that: The preset model building strategy includes the following steps: S10, determining the corresponding model range; S11, creating a plurality of model grids according to geological principles and existing geological data, and setting the grid density corresponding to the model grid based on the position information of the corresponding preset electrode in the corresponding model grid, wherein the grid density gradually becomes sparser as the distance between the grid and the preset electrode increases; S12. Setting different resistivity values for each model network to form each final dielectric resistivity model.
3. A sample data set generator for deep learning inversion of mine resistivity according to claim 2, characterized in that: The preset mine working face construction conditions include: S20, in each of the formed medium resistivity models, determining a geometric model to be set in a grid corresponding to each of the medium resistivity models; S21, generating corresponding two-dimensional resistivity models in batches according to the determined geometric model, and setting corresponding model parameters and the number of two-dimensional resistivity models to be generated; S22, mapping the batch-generated two-dimensional resistivity models onto the medium resistivity model, and setting different resistivities for different parts of the mapped two-dimensional resistivity models to form structural models of the working faces of various mines.
4. A sample data set generator for deep learning inversion of mine resistivity according to claim 3, characterized in that: The preset data acquisition simulation strategy includes the following steps: S30, according to each formed mine working face structure model, discretize the model area corresponding to the mine working face structure model to generate a corresponding grid; S31, setting a corresponding power source and a current characteristic corresponding to the power source, and determining the electrode point position and the current injection point corresponding to the electrode arrangement in the model area based on the characteristics of the model area, and associating the electrode at the current injection point with the power source based on the electrode point position and the current injection point; S32. After the electrode at the current injection point is associated with the power supply, the corresponding power supply is set to the emission current mode, and current is supplied to the underground medium through the electrode associated with the power supply to establish a stable current field. The voltage difference between each electrode point is calculated, and based on the corresponding voltage difference, data related to the resistivity of the underground medium is obtained.
5. A sample data set generator for deep learning inversion of mine resistivity according to claim 4, characterized in that: The processing module is also used to quantify the interference factors according to the interference factors, calculate the corresponding interference values, and add them to the corresponding simulated resistivity observation values.
6. A sample data set generator for deep learning inversion of mine resistivity according to claim 5, characterized in that: The data generation module comprises: A data sorting module is used to identify the labels corresponding to the simulated resistivity observation values according to the formed simulated resistivity observation values and the force resistivity observation values corresponding to the simulated resistivity observation values, wherein the labels include resistivity model parameters, interference factors and noise levels corresponding to the simulated resistivity observation values; A profile generating module is used to generate a corresponding dielectric resistivity profile according to each simulated resistivity observation value; the dielectric resistivity profile is a three-dimensional dielectric resistivity profile or a two-dimensional dielectric resistivity profile; The sample data construction module is used to take the simulated resistivity observation value as the input feature to construct the feature matrix corresponding to the data set, and take the medium resistivity profile corresponding to the simulated resistivity observation value as the output feature to form the corresponding sample data set; The format conversion module is used to convert the formed sample data set into a specified format as the corresponding deep learning training data set.