Method, medium and system for establishing geometric model of overlying strata structure

Abstract

The invention provides an overlying strata structure geometric model establishing method, medium and system, and belongs to the technical field of overlying strata structure geometric models.The method includes the steps that firstly, obtained data are preprocessed, and then three-dimensional grid geometric models based on geological radar, seismic waves and drilling data are established through different data analysis technologies. The three models are integrated by using a complex number neural network fusion method, and an initial integrated three-dimensional grid geometric model is established; optimizing the model by adopting a self-adaptive grid refining technology, and establishing a rock parameter conversion model of a grid unit; and finally, constructing a multi-scale geometric model of the overlying strata structure by adopting a wavelet multi-resolution analysis and Monte Carlo simulation method, improving the model precision through iterative optimization and actual measurement data verification, and finally forming a high-precision three-dimensional geometric model, thereby solving the problem that the uncertainty of the model is difficult to quantify in the existing method, and improving the accuracy of the model. And the risk is difficult to fully assess in engineering application.

Description

Technical Field

[0001] The present invention belongs to the technical field of geometric models of overburden structures, and in particular, relates to a method, medium and system for establishing a geometric model of an overburden structure. Background Art

[0002] Currently, three-dimensional modeling and parametric description of overburden structure is a key technology in geotechnical engineering and geological exploration. Accurate description of the geometric characteristics and mechanical properties of overburden structure is crucial for applications such as geological disaster prediction, engineering construction, and mineral exploration. Traditional overburden structure modeling methods usually rely on a single detection method, such as geological drilling, seismic detection, or geological radar detection. These methods have their own advantages and disadvantages, and it is difficult to fully and accurately characterize the complexity of overburden structure.

[0003] The drilling sampling method commonly used in geological exploration can directly obtain core samples and understand parameters such as lithology, integrity, and RQD, but it can only reflect the overburden conditions in local areas and it is difficult to obtain overall three-dimensional information. The seismic detection method is based on the propagation characteristics of seismic waves in different media for imaging, which can reflect the overall three-dimensional structure of the overburden, but it is difficult to accurately and quantitatively describe complex geological structures. Geological radar detection can obtain the reflection characteristics of electromagnetic waves in the overburden medium, and has a good resolution for shallow overburden structures, but it is difficult to detect deep structures.

[0004] In addition, existing overburden structure modeling methods are usually limited to a single scale and are difficult to reflect the characteristic differences of overburden structures at different scales. Overburden structures have significant scale effects from microscopic particles to overall geological structures, and a single-scale model is difficult to fully characterize its complex characteristics. At the same time, existing methods find it difficult to quantify the uncertainty of the model, making it difficult to fully assess risks in engineering applications. Summary of the invention

[0005] In view of this, the present invention provides a method, medium and system for establishing a geometric model of an overburden structure, which can solve the technical problem that it is difficult to quantify the uncertainty of the model in existing methods, resulting in difficulty in fully assessing risks in engineering applications.

[0006] The present invention is achieved in that:

[0007] A first aspect of the present invention provides a method for establishing a geometric model of an overburden structure, comprising the following steps:

[0008] S10, obtaining detection data of the overburden structure obtained by various detection methods, including geological radar detection data, seismic wave detection data and drilling sampling data;

[0009] S20, preprocessing the acquired detection data;

[0010] S30, using the reflection wave time domain analysis and spectrum analysis method to establish a first three-dimensional grid geometric model of the overburden structure according to the preprocessed geological radar detection data; using travel time tomography and full waveform inversion technology to establish a second three-dimensional grid geometric model of the overburden structure according to the preprocessed seismic wave detection data;

[0011] S40, based on the borehole sampling data, using the Kriging interpolation method to construct a third three-dimensional grid geometric model of the overburden structure, and using it as a constraint condition to optimize the first and second three-dimensional grid geometric models;

[0012] S50, using a complex neural network fusion method considering constraints, integrating information of the first, second and third three-dimensional mesh geometric models to establish an initial integrated three-dimensional mesh geometric model;

[0013] S60, using adaptive mesh refinement technology, locally encrypt and optimize the initial integrated 3D mesh geometry model according to the geological complexity and detection data density;

[0014] S70, based on rock physics relationships, establish a conversion model between the complex equations of the grid unit and the rock mechanics parameters to achieve a comprehensive parameterized description of each grid unit;

[0015] S80. Construct a multi-scale geometric model of the overburden structure, use the wavelet multi-resolution analysis method to decompose the complex equations, obtain the overburden structure characteristics at different scales, and achieve seamless connection between models of different scales;

[0016] S90. Using the Monte Carlo simulation method, the coefficients of the complex equations are randomly perturbed to generate multiple sets of geometric models of the overburden structure. The uncertainty of the model is evaluated, and the accuracy of the model is improved through iterative optimization and measured data verification, ultimately forming a high-precision, multi-scale three-dimensional geometric model of the overburden structure.

[0017] Specifically, the step S10 specifically includes: step 101, obtaining detection data of the overburden structure obtained by various detection means, including geological radar detection data, seismic wave detection data and borehole sampling data; wherein the geological radar detection data includes electromagnetic wave reflection time, waveform amplitude and frequency, the seismic wave detection data includes P wave and S wave velocity, waveform characteristics and attenuation coefficient, and the borehole sampling data includes lithology description, core integrity, RQD value and joint characteristics. Step 102, preprocessing the acquired detection data, including filtering and denoising, data correction and format unification, so as to eliminate noise interference in the detection data, correct systematic errors, and convert data of different formats into a unified data structure.

[0018] Specifically, the specific steps of step S20 include: step 201, using the reflection wave time domain analysis and spectrum analysis method to establish a first three-dimensional grid geometric model of the overburden structure according to the preprocessed geological radar detection data; step 202, using travel time tomography and full waveform inversion technology to establish a second three-dimensional grid geometric model of the overburden structure according to the preprocessed seismic wave detection data; the overburden structure in each grid is represented by a group of complex equations to better describe its complex characteristics.

[0019] Specifically, the specific steps of step S30 include: step 301, based on the drilling sampling data, using the Kriging interpolation method to construct a third three-dimensional grid geometric model of the overburden structure; step 302, using the spatial correlation of known sampling points, and calculating the attribute values ​​of unknown points by the optimal linear unbiased estimation method, thereby obtaining the third three-dimensional grid geometric model; this model can better reflect the actual situation of the overburden structure and provide important constraints for subsequent data fusion.

[0020] Specifically, the specific steps of step S40 include: step 401, using a complex neural network fusion method that takes into account constraint conditions, integrating the information of the first, second and third three-dimensional grid geometric models to establish an initial comprehensive three-dimensional grid geometric model; step 402, the complex neural network can better capture the complex characteristics of the overburden structure, and make full use of the constraint conditions provided by different detection methods for optimization, so as to obtain a more accurate and reliable initial comprehensive three-dimensional grid geometric model.

[0021] Specifically, the specific steps of step S50 include: step 501, using adaptive grid refinement technology, according to the geological complexity and detection data density, the initial comprehensive three-dimensional grid geometric model is locally encrypted and optimized to obtain an optimized three-dimensional grid geometric model; step 502, this technology can refine and optimize complex areas on the basis of retaining the overall structure of the model, so as to further improve the accuracy and computational efficiency of the model.

[0022] Specifically, the specific steps of step S60 include: step 601, based on rock physics relationships, establishing a conversion model between the complex equations of the grid unit and the rock mechanics parameters, to achieve a comprehensive parameterized description of each grid unit, including elastic modulus, Poisson's ratio and compressive strength; step 602, the parameterized description can provide an important basis for subsequent rock mechanics analysis and engineering applications.

[0023] Specifically, the specific steps of step S70 include: step 701, using wavelet multi-resolution analysis method to decompose the complex equation group to obtain the overburden structure characteristics at different scales; step 702, combining detail coefficients and approximate coefficients of different scales to achieve seamless connection between models of different scales; this multi-scale decomposition method can more comprehensively describe the characteristics of the overburden structure and provide a basis for subsequent uncertainty assessment and model optimization.

[0024] Specifically, the specific steps of step S80 include: step 801, using the Monte Carlo simulation method to randomly perturb the coefficients of the complex equation group to generate multiple groups of overburden structure geometric models; step 802, evaluating the uncertainty index of the generated model, and further improving the accuracy of the model through iterative optimization to make it better match the measured data; this method can comprehensively evaluate the uncertainty of the model and continuously optimize the model, and finally form a high-precision, multi-scale three-dimensional geometric model of the overburden structure.

[0025] Furthermore, the first three-dimensional grid geometric model of the overburden structure is established based on the preprocessed geological radar detection data using the reflected wave time domain analysis and spectrum analysis method, which can be expressed as:

[0026] M1(x,y,z)=f(t(x,y,z),A(x,y,z),ω(x,y,z));

[0027] Where M1(x, y, z) is the first three-dimensional mesh geometry model; x, y, z are spatial coordinates; t(x, y, z) is the electromagnetic wave reflection time; A(x, y, z) is the waveform amplitude; ω(x, y, z) is the frequency; and f is the function of time domain analysis and spectrum analysis of the reflected wave.

[0028] Furthermore, using travel-time tomography and full waveform inversion technology, a second three-dimensional grid geometric model of the overburden structure is established based on the preprocessed seismic wave detection data, which can be expressed as:

[0029] M2(x,y,z)=g(V p (x,y,z),V s (x,y,z),W(x,y,z),α(x,y,z));

[0030] Where M2(x, y, z) is the second three-dimensional mesh geometry model; V p (x, y, z) is the P wave velocity; V s (x, y, z) is the S-wave velocity; W(x, y, z) is the waveform characteristic; α(x, y, z) is the attenuation coefficient; g is the function of travel-time tomography and full waveform inversion.

[0031] Furthermore, the overburden structure in each grid is represented by a set of complex equations, which can be expressed as:

[0032] C(x,y,z)=a(x,y,z)+ib(x,y,z);

[0033] Where C(x,y,z) is a complex equation system; a(x,y,z) is the real part; b(x,y,z) is the imaginary part; and i is the imaginary unit.

[0034] Furthermore, the Kriging interpolation method is used to construct a third 3D grid geometry model:

[0035]

[0036] Where M3(x,y,z) is the third three-dimensional mesh geometry model; Z(x i ,y i ,z i ) is the observed value of the known drilling sampling point; i is the weight coefficient, satisfying n is the number of known drilling sampling points.

[0037] Weight coefficient λ i The calculation of is obtained by solving the following system of equations:

[0038]

[0039] In the formula, γ(h ij ) is the semivariogram, which represents the spatial correlation between two sampling points; h ij is the distance between the i-th and j-th sampling points; μ is the Lagrange multiplier; h i0 is the distance between the i-th sampling point and the point to be estimated.

[0040] Furthermore, the initial comprehensive three-dimensional mesh geometry model is established by using a complex neural network fusion method considering constraints:

[0041] M0(x,y,z)=NN(M1(x,y,z),M2(x,y,z),M3(x,y,z));

[0042] Where M0(x, y, z) is the initial comprehensive three-dimensional mesh geometry model; NN is the complex neural network function.

[0043] The input layer of the complex neural network can be expressed as:

[0044] I=[M1(x,y,z),M2(x,y,z),M3(x,y,z)];

[0045] The calculation process of the hidden layer can be expressed as:

[0046]

[0047] In the formula, H j is the output of the jth hidden layer neuron; σ is the activation function; w jk is the weight from the input layer to the hidden layer; b j is the bias term.

[0048] The calculation process of the output layer can be expressed as:

[0049]

[0050] Where O is the result of the output layer, i.e., M0(x,y,z); v j is the weight from the hidden layer to the output layer; c is the bias term; m is the number of neurons in the hidden layer.

[0051] Furthermore, the adaptive mesh refinement technology is used to locally encrypt and optimize the initial integrated 3D mesh geometry model:

[0052]

[0053] Where M r (x, y, z) is the optimized three-dimensional mesh geometry model; R is the adaptive mesh refinement function; is the gradient of the initial model; ρ(x,y,z) is the detection data density.

[0054] Furthermore, a conversion model between the complex equations of the grid unit and the rock mechanics parameters is established:

[0055]

[0056] Where E(x,y,z) is the elastic modulus; v(x,y,z) is the Poisson's ratio; σ c (x,y,z) is the compressive strength; Re(C(x,y,z)) and Im(C(x,y,z)) are the real and imaginary parts of the complex equation system respectively; f E ,f v ,f σ is the corresponding conversion function.

[0057] Furthermore, the wavelet multi-resolution analysis method is used to decompose the complex equations:

[0058]

[0059] Where D j (x, y, z) is the detail coefficient of the jth scale; A J(x, y, z) are the approximation coefficients of the Jth scale; J is the maximum scale of the decomposition.

[0060] The calculation of detail coefficient and approximation coefficient can be expressed as:

[0061] D j (x,y,z)=∑ k ∑ l ∑ m d j,k,l,m ψ j,k,l,m (x,y,z);

[0062] A J (x,y,z)=∑ k ∑ l ∑ m a J,k,l,m φ J,k,l,m (x,y,z);

[0063] Where, d j,k,l,m and a J,k,l,m are wavelet coefficients and scaling coefficients respectively; ψ j,k,l,m (x,y,z) and φ J,k,l,m (x, y, z) are the wavelet function and scaling function respectively.

[0064] Furthermore, the Monte Carlo simulation method is used to randomly perturb the coefficients of the complex equations:

[0065] C s (x,y,z)=C(x,y,z)+δ(x,y,z);

[0066] In the formula, C s (x, y, z) is the perturbed complex equation system; δ(x, y, z) is a random perturbation term that obeys a certain probability distribution (such as normal distribution).

[0067] Generate multiple sets of overburden structure geometry models:

[0068] M s (x,y,z)=h(C s (x,y,z));

[0069] Where M s (x, y, z) is the generated geometric model of the overburden structure; h is the mapping function from the complex equations to the geometric model.

[0070] Model uncertainty assessment:

[0071]

[0072] Where U is the uncertainty index; N is the number of generated models; is the i-th generated model; is the average of all generated models.

[0073] The iterative optimization process can be expressed as:

[0074]

[0075] Where M k (x, y, z) is the model of the kth iteration; α is the learning rate; is the gradient of the loss function with respect to the model.

[0076] The loss function can be defined as the error between the model prediction and the measured data:

[0077] L(M k (x,y,z))=||M k (x,y,z)-D(z,y,z)|| 2 ;

[0078] Where D(x,y,z) is the measured data.

[0079] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for establishing a geometric model of an overburden structure.

[0080] A third aspect of the present invention provides a system for establishing a geometric model of an overburden structure, which includes the above-mentioned computer-readable storage medium.

[0081] Compared with the prior art, the method, medium and system for establishing a geometric model of overburden structure provided by the present invention can fully integrate the information of different detection methods, construct a high-precision, multi-scale three-dimensional model of overburden structure, and quantitatively analyze the uncertainty of the model, providing reliable support for subsequent engineering applications. The beneficial effects of the present invention are:

[0082] 1. Integrate multi-source detection data and make full use of the advantages of different detection methods. Geological radar detection can obtain high-resolution three-dimensional information of the overburden structure, seismic wave detection can reflect the overall three-dimensional structural characteristics, and borehole sampling provides the actual parameters of the overburden. The method of the present invention can effectively integrate this information to construct a more accurate and reliable overburden structure model.

[0083] 2. Adopt adaptive grid refinement technology to perform local optimization for complex areas. Conventional regular grids are difficult to fully describe the complexity of overburden structure. The method of the present invention adaptively adjusts the grid density according to geological conditions and detection data density, increases the number of grids in complex areas, and reduces the number of grids in simple areas, thereby improving the accuracy of the model.

[0084] 3. Establish a conversion model between complex equations and rock mechanics parameters to achieve a comprehensive parametric description. Traditional methods can only describe geometric features and are difficult to associate with the mechanical parameters required for engineering applications. The method of the present invention can provide important support for subsequent rock mechanics analysis and engineering design by establishing a conversion relationship between complex equations and parameters such as elastic modulus, Poisson's ratio, and compressive strength.

[0085] 4. Use wavelet multi-resolution analysis and Monte Carlo simulation methods to achieve multi-scale representation and uncertainty analysis. There is a significant scale effect in the overburden structure. The method of the present invention can extract characteristic information at different scales through wavelet analysis, and use Monte Carlo simulation to evaluate the uncertainty of the model, providing more comprehensive and reliable information support for engineering applications.

[0086] In general, the method of the present invention integrates a variety of advanced data processing and modeling technologies, which can more comprehensively and accurately describe the geometric characteristics and mechanical properties of the overburden structure, provide reliable technical support for the fields of geotechnical engineering, geological exploration, etc., and solve the technical problem that it is difficult to quantify the uncertainty of the model in existing methods, resulting in difficulty in fully assessing risks in engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0089] like Figure 1 FIG. 1 is a flow chart of a method for establishing a geometric model of an overburden structure provided by the first aspect of the present invention. The method comprises the following steps:

[0090] S10, obtaining detection data of the overburden structure obtained by various detection means, including geological radar detection data, seismic wave detection data and borehole sampling data, wherein the geological radar detection data includes electromagnetic wave reflection time, waveform amplitude and frequency; the seismic wave detection data includes P-wave and S-wave velocity, waveform characteristics and attenuation coefficient; the borehole sampling data includes lithology description, core integrity, RQD value and joint characteristics;

[0091] S20, preprocessing the acquired detection data, including filtering and denoising, data correction and format unification;

[0092] S30, using the reflection wave time domain analysis and spectrum analysis method, establishing a first three-dimensional grid geometric model of the overburden structure according to the preprocessed geological radar detection data; using travel time tomography and full waveform inversion technology, establishing a second three-dimensional grid geometric model of the overburden structure according to the preprocessed seismic wave detection data; the first three-dimensional grid geometric model and the second three-dimensional grid geometric model use grids of the same size, and the overburden structure in each grid uses a complex equation group to represent the overburden structure in most grids, and the first three-dimensional grid geometric model is a three-dimensional matrix, and the elements of each matrix are complex equation groups corresponding to coordinates;

[0093] S40, based on the borehole sampling data, using the Kriging interpolation method to construct a third three-dimensional grid geometric model of the overburden structure, and using it as a constraint condition to optimize the first and second three-dimensional grid geometric models;

[0094] S50, using a complex neural network fusion method considering constraints, integrating information of the first, second and third three-dimensional mesh geometric models to establish an initial integrated three-dimensional mesh geometric model;

[0095] S60, using adaptive mesh refinement technology, locally encrypt and optimize the initial integrated 3D mesh geometry model according to the geological complexity and detection data density;

[0096] S70. Based on rock physics relationships, a conversion model between the complex equations of the grid unit and the rock mechanics parameters is established to achieve a comprehensive parameterized description of each grid unit, including elastic modulus, Poisson's ratio and compressive strength;

[0097] S80. Construct a multi-scale geometric model of the overburden structure, use the wavelet multi-resolution analysis method to decompose the complex equations, obtain the overburden structure characteristics at different scales, and achieve seamless connection between models of different scales;

[0098] S90. Using the Monte Carlo simulation method, the coefficients of the complex equations are randomly perturbed to generate multiple sets of geometric models of the overburden structure. The uncertainty of the model is evaluated, and the accuracy of the model is improved through iterative optimization and measured data verification, ultimately forming a high-precision, multi-scale three-dimensional geometric model of the overburden structure.

[0099] The specific implementation methods of the above steps are described in detail below:

[0100] The specific implementation of step S10 is as follows:

[0101] First, the detection data of the overburden structure obtained by various detection methods are obtained, including geological radar detection data, seismic wave detection data and drilling sampling data. The geological radar detection data includes the electromagnetic wave reflection time t(x,y,z), waveform amplitude A(x,y,z) and frequency ω(x,y,z); the seismic wave detection data includes the P wave velocity V p (x,y,z), S wave velocity V s (x, y, z), waveform characteristics W(x, y, z) and attenuation coefficient α(x, y, z); drilling sampling data include lithology description, core integrity, RQD value and joint characteristics Z(x i ,y i ,z i ), where x i ,y i ,z i Represents the spatial coordinates of the i-th drilling sampling point.

[0102] Since these detection data come from different measurement equipment and methods, they need to be preprocessed to unify the data format. Preprocessing includes filtering and denoising, data correction and format unification. Filtering and denoising can eliminate noise interference in the detection data and improve the signal-to-noise ratio (SNR) of the data; data correction can correct the systematic error ε caused by instrument errors, environmental factors, etc., to ensure the accuracy of the data; format unification can convert detection data in different formats into a unified data structure, providing a basis for subsequent data fusion and modeling.

[0103] Specifically, filtering and denoising can be performed using the following methods:

[0104]

[0105] Among them, P signal is the signal power, P noise is the noise power. By designing a suitable filter H(f), the signal-to-noise ratio SNR can be improved to obtain clean detection data.

[0106] Data correction can be done by:

[0107] Z corrected =Z observed -ε;

[0108] Among them, Z observed is the observed value, and ε is the systematic error. By subtracting the systematic error ε, the corrected data Z can be obtained. corrected .

[0109] Format unification can convert detection data of different formats into a format that can be directly input into subsequent algorithms by defining a unified data structure. For example, geological radar detection data, seismic wave detection data, and borehole sampling data can be uniformly converted into three-dimensional grid data, and each grid unit contains the corresponding physical quantity.

[0110] Through the above preprocessing steps, clean and uniformly formatted overburden structure detection data were obtained, laying the foundation for the subsequent three-dimensional model establishment.

[0111] The specific implementation of step S20 is as follows:

[0112] The first three-dimensional grid geometric model M1(x, y, z) of the overburden structure is established based on the preprocessed geological radar detection data using the reflection wave time domain analysis and spectrum analysis method. Reflection wave time domain analysis and spectrum analysis is a commonly used geological radar data processing method. It uses the reflection characteristics of electromagnetic waves in the overburden medium to obtain the three-dimensional spatial distribution information of the overburden structure by analyzing the reflection time t(x, y, z), amplitude A(x, y, z) and frequency ω(x, y, z). Specifically, the first three-dimensional grid geometric model M1(x, y, z) can be expressed as:

[0113] M1(x,y,z)=f(t(x,y,z),A(x,y,z),ω(x,y,z));

[0114] Among them, f is the function of time domain analysis and spectrum analysis of reflected waves. This model can better describe the three-dimensional geometric characteristics of the overburden structure.

[0115] At the same time, travel-time tomography and full waveform inversion techniques are used to establish a second three-dimensional grid geometric model M2 (x, y, z) of the overburden structure based on the preprocessed seismic wave detection data. Travel-time tomography is a three-dimensional imaging method based on the propagation velocity of seismic waves, which can reflect the P-wave and S-wave velocity distribution V of the overburden structure. p (x,y,z) and V s (x, y, z); full waveform inversion can use the complete waveform information W(x, y, z) and attenuation coefficient α(x, y, z) of the seismic wave to more accurately restore the three-dimensional characteristics of the overburden structure. The second three-dimensional grid geometric model can be expressed as:

[0116] M2(x,y,z)=g(V p (x,y,z),V s (x,y,z),W(x,y,z),α(x,y,z));

[0117] Here, g is a function of travel-time tomography and full waveform inversion.

[0118] It should be noted that the first and second 3D mesh geometry models use meshes of the same size, and the overburden structure in each mesh is represented by the complex equation system C(x,y,z)=a(x,y,z)+ib(x,y,z). The advantage of this is that it can better describe the complex characteristics of the overburden structure and lay the foundation for subsequent multi-source data fusion.

[0119] The specific implementation of step S30 is as follows:

[0120] Based on the borehole sampling data, the Kriging interpolation method is used to construct the third 3D grid geometric model M3(x, y, z) of the overburden structure. The Kriging interpolation method is a commonly used geostatistical interpolation method. It uses the spatial correlation of known sampling points and the optimal linear unbiased estimation method to calculate the attribute values ​​of unknown points. Specifically, the third 3D grid geometric model can be expressed as:

[0121]

[0122] Among them, Z(x i ,y i ,z i ) is the observed value of the known drilling sampling point, λ i is the weight coefficient, satisfying Weight coefficient λ i The calculation of is obtained by solving the semivariogram equation system:

[0123]

[0124] Among them, γ(h ij ) is the semivariogram, which represents the spatial correlation between two sampling points; h ij is the distance between the i-th and j-th sampling points; μ is the Lagrange multiplier; h i0 is the distance between the i-th sampling point and the point to be estimated.

[0125] The third three-dimensional grid geometric model M3 (x, y, z) can better reflect the actual situation of the overburden structure and provide important constraints for subsequent data fusion.

[0126] The specific implementation of step S40 is as follows:

[0127] The complex neural network fusion method considering the constraints is used to integrate the information of the first, second and third three-dimensional mesh geometric models to establish the initial comprehensive three-dimensional mesh geometric model M0 (x, y, z). The complex neural network is a neural network model that can process complex inputs and outputs, which can better capture the complex characteristics of the overburden structure. Specifically, the input layer of the complex neural network can be expressed as:

[0128] I=[M1(x,y,z),M2(x,y,z),M3(x,y,z)];

[0129] The calculation process of the hidden layer can be expressed as:

[0130]

[0131] Among them, H j is the output of the jth hidden layer neuron, σ is the activation function, w jk is the weight from the input layer to the hidden layer, b j is the bias term.

[0132] The calculation process of the output layer can be expressed as:

[0133]

[0134] Among them, O is the result of the output layer, that is, M0(x,y,z), v j is the weight from the hidden layer to the output layer, c is the bias term, and m is the number of neurons in the hidden layer.

[0135] This fusion method can comprehensively utilize the information provided by different detection methods and make full use of the constraints for optimization to obtain a more accurate and reliable initial comprehensive three-dimensional mesh geometric model M0 (x, y, z).

[0136] The specific implementation of step S50 is as follows:

[0137] Using the adaptive mesh refinement technology, the initial comprehensive 3D mesh geometric model M0(x, y, z) is locally encrypted and optimized according to the geological complexity and detection data density to obtain the optimized 3D mesh geometric model M r (x,y,z). Adaptive mesh refinement technology can be used based on the gradient information of the model. and detection data density ρ(x,y,z), increase the mesh density in complex areas and reduce the mesh density in simple areas, thereby improving the accuracy and computational efficiency of the model. Specifically, the optimized 3D mesh geometry model can be expressed as:

[0138]

[0139] Among them, R is the adaptive mesh refinement function. Through this method, on the basis of retaining the overall structure of the model, the complex area can be refined and optimized to obtain a more refined and accurate three-dimensional geometric model of the overburden structure.

[0140] The specific implementation of step S60 is as follows:

[0141] Based on the rock physics relationship, a conversion model between the complex equations of the grid unit and the rock mechanics parameters is established to achieve a comprehensive parameterized description of each grid unit. Specifically, the following conversion model can be established:

[0142]

[0143] Where E(x,y,z) is the elastic modulus, v(x,y,z) is the Poisson's ratio, σ c (x,y,z) is the compressive strength, Re(C(x,y,z)) and Im(C(x,y,z)) are the real and imaginary parts of the complex equation system, respectively, f E ,f v ,f σ is the corresponding conversion function. This parametric description can provide an important basis for subsequent rock mechanics analysis and engineering applications.

[0144] The specific implementation of step S70 is as follows:

[0145] The wavelet multi-resolution analysis method is used to decompose the complex equation group C(x, y, z) to obtain the structural characteristics of the overburden at different scales and achieve seamless connection between models of different scales. Wavelet analysis is a multi-resolution analysis method that can decompose the signal into detail coefficients and approximate coefficients of different scales. Specifically, the decomposition process of the complex equation group can be expressed as:

[0146]

[0147] Among them, D j (x, y, z) is the detail coefficient of the jth scale, A J (x, y, z) is the approximate coefficient of the Jth scale, and J is the maximum scale of decomposition. The calculation of detail coefficient and approximate coefficient can be expressed as:

[0148] D j (x,y,z)=∑ k ∑ l ∑ m d j,k,l,m ψ j,k,l,m (x,y,z);

[0149] A J (x,y,z)=∑ k ∑ l ∑ m a J,k,l,m φ J,k,l,m (x,y,z);

[0150] Among them, d j,k,l,m and a J,k,l,m are wavelet coefficients and scaling coefficients, respectively, j,k,l,m(x,y,z) and φ J,k,l,m (x, y, z) are the wavelet function and scaling function respectively.

[0151] This multi-scale decomposition method can more comprehensively describe the characteristics of the overburden structure and provide a basis for subsequent uncertainty assessment and model optimization.

[0152] The specific implementation of step S80 is as follows:

[0153] The Monte Carlo simulation method is used to randomly perturb the coefficients of the complex equations to generate multiple sets of overburden structure geometric models and evaluate the uncertainty of the model. Specifically, the perturbed complex equations can be expressed as:

[0154] C s (x,y,z)=C(x,y,z)+δ(x,y,z);

[0155] Among them, δ(x, y, z) is a random perturbation term that obeys a certain probability distribution (such as normal distribution). Based on the perturbed complex equation system C s (x, y, z), multiple sets of overburden structure geometric models M can be generated s (x,y,z):

[0156] M s (x,y,z)=h(C s (x,y,z));

[0157] Where h is the mapping function from the complex equations to the geometric model.

[0158] By generating multiple sets of models, the uncertainty index U of the model can be evaluated:

[0159]

[0160] Where N is the number of generated models, is the i-th generated model, is the average of all generated models.

[0161] On this basis, the accuracy of the model can be further improved through iterative optimization to better match the measured data. The iterative optimization process can be expressed as:

[0162]

[0163] Among them, M k (x, y, z) is the model of the kth iteration, α is the learning rate, is the gradient of the loss function with respect to the model. The loss function can be defined as the error between the model prediction value and the measured data:

[0164] L(M k (x,y,z))=||M k (x,y,z)-D(x,y,z)|| 2 ;

[0165] Among them, D(x,y,z) is the measured data.

[0166] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for establishing a geometric model of an overburden structure.

[0167] A third aspect of the present invention provides a system for establishing a geometric model of an overburden structure, which includes the above-mentioned computer-readable storage medium.

[0168] Specifically, the principle of the present invention is:

[0169] 1. Make full use of the complementarity of multi-source detection data. Geological radar detection, seismic wave detection and borehole sampling each have their own focus. Geological radar is good at high-resolution imaging of shallow structures, seismic wave detection can reflect the overall three-dimensional structural characteristics, and borehole sampling provides the actual parameters of the rock mass. The method of the present invention integrates these detection data, using their respective advantages to make up for each other's shortcomings, thereby constructing a more comprehensive and reliable overburden structure model.

[0170] 2. Adopt adaptive grid refinement technology for local optimization. The overburden structure has significant spatial heterogeneity, and it is difficult for conventional regular grids to fully capture its complex features. The method of the present invention adaptively adjusts the grid density according to the geological complexity and the density of the detection data, increases the number of grids in complex areas, and reduces the number of grids in simple areas, thereby highlighting the local detail information while retaining the overall structural characteristics. This adaptive grid refinement method can improve the accuracy and computational efficiency of the model.

[0171] 3. Establish a conversion model between complex equations and rock mechanics parameters. Traditional geometric models are difficult to directly associate with the mechanical parameters required for practical engineering applications. The method of the present invention can achieve a comprehensive parametric description of each grid unit by establishing a conversion relationship between complex equations and parameters such as elastic modulus, Poisson's ratio, and compressive strength, providing an important basis for subsequent rock mechanics analysis and engineering design.

[0172] 4. Use wavelet multi-resolution analysis to achieve multi-scale representation. There is a significant scale effect in the overburden structure, and there are characteristic differences at multiple scales from microscopic particles to the overall geological structure. The method of the present invention uses wavelet analysis technology to decompose the complex equation system at multiple scales and extract characteristic information at different scales. This multi-scale representation can more comprehensively describe the complex characteristics of the overburden structure and provide a basis for uncertainty analysis and optimization.

[0173] 5. Uncertainty analysis combined with Monte Carlo simulation. Due to the geological complexity of the overburden structure itself and the limitations of the detection data, the model has inevitable uncertainty. The method of the present invention adopts the Monte Carlo simulation method to generate multiple groups of overburden structure geometric models by randomly perturbing the coefficients of the complex equation group, and quantitatively evaluates their uncertainty indicators. This method can provide more comprehensive and reliable risk assessment information for engineering applications.

[0174] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: a mining area is located in a complex mountain geological environment, and the overburden structure has obvious spatial heterogeneity and scale effect. In order to ensure the safety and reliability of engineering construction, detailed exploration and modeling analysis of the overburden structure is required.

[0175] Data acquisition and preprocessing

[0176] First, comprehensive geological exploration work was carried out at the mining site, including geological radar exploration, seismic wave exploration and drill sampling.

[0177] The geological radar exploration uses electromagnetic wave detection equipment with a frequency of 100MHz, and 300 measuring points are set up on the site with a spacing of 10 meters between the measuring points. Through time domain analysis and spectrum analysis, the electromagnetic wave reflection time t(x,y,z), amplitude A(x,y,z) and frequency ω(x,y,z) data of the overburden structure are obtained. The seismic wave exploration uses P-wave and S-wave detection technology, and 200 measuring points are set up on the site with a spacing of 20 meters between the measuring points. Through travel time tomography and full waveform inversion analysis, the P-wave velocity V of the overburden structure is obtained. p (x,y,z), S wave velocity V s (x, y, z), waveform characteristics W(x, y, z) and attenuation coefficient α(x, y, z). In addition, 10 drilling sampling points were arranged on the site to obtain data such as lithology description of overburden, core integrity, RQD value and joint characteristics Z(x i ,y i ,z i ).

[0178] In order to achieve unified data processing, the acquired detection data was first preprocessed. The geological radar detection data was denoised by a bandpass filter to eliminate the influence of high-frequency noise on the time resolution and improve the signal-to-noise ratio (SNR). The seismic wave detection data was corrected by the instrument error ε to ensure the accuracy of the data. Finally, the detection data in different formats were uniformly converted into a three-dimensional grid structure, and each grid unit contained the corresponding physical quantity.

[0179] Building 3D mesh geometry models

[0180] Based on the preprocessed detection data, the method of the present invention constructs three three-dimensional grid geometric models respectively.

[0181] Firstly, the first three-dimensional grid geometric model M1(x,y,z) of the overburden structure was established based on the geological radar detection data using the reflected wave time domain analysis and spectrum analysis methods. Specifically, the model can be expressed as:

[0182] M1(x,y,z)=f(t(x,y,z),A(x,y,z),ω(x,y,z));

[0183] Among them, f is the function of time domain analysis and spectrum analysis of reflected waves. This model can better describe the three-dimensional geometric characteristics of shallow overburden structure.

[0184] Then, using travel time tomography and full waveform inversion technology, a second three-dimensional grid geometric model M2 (x, y, z) of the overburden structure was established based on the seismic wave detection data. The model can be expressed as:

[0185] M2(x,y,z)=g(V p (x,y,z),V s (x,y,z),W(x,y,z),α(x,y,z));

[0186] Where g is a function of travel-time tomography and full waveform inversion. The model can reflect the overall three-dimensional characteristics of the overburden structure.

[0187] Finally, based on the borehole sampling data, the third three-dimensional grid geometric model M3 (x, y, z) of the overburden structure was constructed using the Kriging interpolation method. The model can be expressed as:

[0188]

[0189] Among them, λ i is the weight coefficient, Z(x i ,y i ,z i ) is the observed value of the drilling sampling point. The model can better reflect the actual situation of the overburden structure.

[0190] In order to facilitate the subsequent data fusion, the grid sizes of the three models are kept consistent, and the overburden structure in each grid unit is expressed by the complex equation system C(x, y, z) = a(x, y, z) + ib(x, y, z).

[0191] Data fusion and model optimization

[0192] Based on the above three three-dimensional mesh geometric models, the method of the present invention adopts the complex neural network fusion technology considering the constraints to construct the initial comprehensive three-dimensional mesh geometric model M0 (x, y, z). The input layer of the complex neural network is the information I = [M1, M2, M3] of the three models, and the output layer O = M0 is obtained after the complex calculation of the hidden layer. This fusion method can comprehensively utilize the information provided by different detection methods and make full use of the constraints for optimization.

[0193] On this basis, the adaptive mesh refinement technology is further used to locally optimize the initial model M0(x, y, z) to obtain the optimized three-dimensional mesh geometric model M r (x,y,z). This technique uses the gradient information of the model and detection data density ρ(x,y,z), increasing the mesh density in complex areas and reducing the mesh density in simple areas, thereby improving the accuracy and computational efficiency of the model.

[0194] In order to achieve a comprehensive parametric description of the overburden structure, the method of the present invention also establishes a conversion model between the complex equations and rock mechanics parameters, including elastic modulus E(x, y, z), Poisson's ratio v(x, y, z) and compressive strength σ c (x,y,z). The conversion relationship can be expressed as:

[0195]

[0196] Among them, f E ,f v ,f σ is the corresponding conversion function. This parametric description can provide an important basis for subsequent rock mechanics analysis and engineering applications.

[0197] Multi-scale feature extraction and uncertainty analysis

[0198] In order to more comprehensively describe the complex characteristics of the overburden structure, the method of the present invention uses wavelet multi-resolution analysis technology to decompose the complex equation group C(x, y, z) and extract characteristic information at different scales. Specifically, the complex equation group can be expressed as:

[0199]

[0200] Among them, Dj (x, y, z) is the detail coefficient of the jth scale, A J (x, y, z) is the approximate coefficient of the Jth scale. This multi-scale decomposition method can more comprehensively describe the characteristic differences of the overburden structure from micro to macro.

[0201] In order to evaluate the uncertainty of the model, the method of the present invention uses the Monte Carlo simulation method to randomly perturb the coefficients of the complex equations and generates 100 sets of overburden structure geometric models M. s (x,y,z). These models can be expressed as:

[0202] C s (x,y,z)=C(x,y,z)+δ(x,y,z);

[0203] M s (x,y,z)=h(C s (x,y,z));

[0204] Among them, δ(x, y, z) is a random disturbance term that obeys the normal distribution, and h is the mapping function from the complex equations to the geometric model.

[0205] By calculating the uncertainty index U of these 100 sets of models, we get:

[0206]

[0207] in, is the average value of all models. This shows that the three-dimensional model of overburden structure constructed by the method of the present invention has lower uncertainty.

[0208] Model validation based on measured data

[0209] In order to further improve the accuracy and reliability of the model, the method of the present invention adopts an iterative optimization method to compare and calibrate the model prediction value with the measured data. Specifically, the iterative optimization process can be expressed as:

[0210]

[0211] Among them, M k (x, y, z) is the model of the kth iteration, α is the learning rate, is the gradient of the loss function with respect to the model. The loss function L can be defined as the error between the model prediction value and the measured data D(x,y,z):

[0212] L(M k (x,y,z))=||M k (x,y,z)-D(x,y,z)|| 2 ;

[0213] Through iterative optimization, the final high-precision, multi-scale 3D geometric model of overburden structure M was obtained. r (x,y,z). This model can accurately reflect the structural characteristics of the overburden at the mining site.

[0214] The model validation results are summarized in Table 1 below:

[0215] Table 1 Model validation results

[0216] parameter Model prediction value Measured data error Elastic modulus E(GPa) 11.2 10.8 3.7％ Poisson's ratio 0.22 0.21 4.8％ <![CDATA[Compressive strength σ c (MPa)]]> 45.3 43.6 3.9％

[0217] It can be seen from the table that the three-dimensional model of the overburden structure constructed by the method of the present invention is in good agreement with the measured data, and the prediction error of each parameter is controlled within 5%, which meets the accuracy requirements of engineering applications.

[0218] Based on the high-precision, multi-scale 3D model of the overburden structure, the rock mechanics analysis and engineering design of the mining site were further carried out. By combining the rock mechanics model, the stress distribution, deformation characteristics and stability index of the overburden structure under engineering loads were predicted. These analysis results provide important decision-making support for the excavation of foundation pits, slope support, foundation design, etc. in the mining area.

[0219] In general, the method of the present invention has achieved good results in the practice of overburden structure modeling at the engineering site of the mining area. The method can make full use of multi-source detection data, construct a high-precision, multi-scale three-dimensional model of the overburden structure, and quantitatively evaluate the uncertainty of the model. This provides reliable technical support for geological survey and engineering design of overburden structure.

[0220] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for establishing a geometric model of an overburden structure, characterized in that: The following steps are involved: S10, obtaining detection data of the overburden structure obtained by various detection methods, including geological radar detection data, seismic wave detection data and drilling sampling data; S20, preprocessing the acquired detection data; S30, using the reflection wave time domain analysis and spectrum analysis method to establish a first three-dimensional grid geometric model of the overburden structure according to the preprocessed geological radar detection data; using travel time tomography and full waveform inversion technology to establish a second three-dimensional grid geometric model of the overburden structure according to the preprocessed seismic wave detection data; S40, based on the borehole sampling data, using the Kriging interpolation method to construct a third three-dimensional grid geometric model of the overburden structure, and using it as a constraint condition to optimize the first and second three-dimensional grid geometric models; S50, using a complex neural network fusion method considering constraints, integrating information of the first, second and third three-dimensional mesh geometric models to establish an initial integrated three-dimensional mesh geometric model; S60, using adaptive mesh refinement technology, locally encrypt and optimize the initial integrated 3D mesh geometry model according to the geological complexity and detection data density; S70, based on rock physics relationships, establish a conversion model between the complex equations of the grid unit and the rock mechanics parameters to achieve a comprehensive parameterized description of each grid unit; S80. Construct a multi-scale geometric model of the overburden structure, use the wavelet multi-resolution analysis method to decompose the complex equations, obtain the overburden structure characteristics at different scales, and achieve seamless connection between models of different scales; S90. Using the Monte Carlo simulation method, the coefficients of the complex equations are randomly perturbed to generate multiple sets of geometric models of the overburden structure. The uncertainty of the model is evaluated, and the accuracy of the model is improved through iterative optimization and measured data verification, ultimately forming a high-precision, multi-scale three-dimensional geometric model of the overburden structure.

2. A method for establishing a geometric model of a covering rock structure according to claim 1, characterized in that: The geological radar detection data includes electromagnetic wave reflection time, waveform amplitude and frequency.

3. A method for establishing a geometric model of a covering rock structure according to claim 2, characterized in that: The seismic wave detection data includes P-wave and S-wave velocities, waveform characteristics and attenuation coefficients.

4. A method for establishing a geometric model of a covering rock structure according to claim 3, characterized in that: The drilling sampling data includes lithology description, core integrity, RQD value and joint characteristics.

5. A method for establishing a geometric model of a covering rock structure according to claim 4, characterized in that: The preprocessing includes filtering and denoising, data correction and format unification.

6. A method for establishing a geometric model of a covering rock structure according to claim 5, characterized in that: The first three-dimensional grid geometric model and the second three-dimensional grid geometric model use grids of the same size, and the overburden structure in each grid uses a complex equation group to represent the overburden structure in most grids.

7. A method for establishing a geometric model of a covering rock structure according to claim 6, characterized in that: The first three-dimensional grid geometric model is a three-dimensional matrix, and each element of the matrix is ​​a set of complex equations corresponding to coordinates.

8. A method for establishing a geometric model of a covering rock structure according to claim 7, characterized in that: A comprehensive parameterized description of each mesh element, including elastic modulus, Poisson's ratio, and compressive strength.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the method for establishing a geometric model of an overburden structure according to any one of claims 1 to 8.

10. A system for establishing a geometric model of a covering rock structure, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images