A method for quantitatively evaluating construction deformation strength based on discrete elements and neural networks

Abstract

This invention belongs to the field of geological structural deformation evaluation technology, specifically involving a quantitative evaluation method for structural deformation intensity based on discrete element method (DEM) and neural network. This invention achieves real seismic profile mapping through parameter combination design, DEM simulation and particle displacement sample construction, neural network displacement response prediction, Q-value quantitative characterization, and K-value normalization grading. Ultimately, it enables the classification of structural zone deformation intensity based on evaluation points or segments of the real seismic profile. This invention eliminates the arbitrariness of manually subjectively setting parameter weights, improves the objectivity of the grading evaluation, and effectively prevents evaluation deviations caused by benchmark offsets when evaluating different models or profiles by using a globally fixed dimensionless benchmark and a solidified background reference sample set. It can be directly applied to real seismic profiles and is suitable for complex structural areas with multiple detachment layers, such as fold-thrust belts and foreland basins. It can provide reliable technical support for oil and gas exploration target selection and engineering risk assessment.

Description

Technical Field

[0001] This invention belongs to the field of geological structural deformation evaluation technology, specifically relating to a quantitative evaluation method for structural deformation intensity based on discrete element method and neural network. Background Technology

[0002] Tectonic deformation intensity refers to the degree of deformation response of strata or granular units in a tectonic profile relative to their initial state under the action of tectonic movement, including vertical displacement, bending, shortening, uplift, subsidence, or fracturing. Existing technologies for the identification and delineation of tectonic deformation zones mainly include qualitative identification methods based on tectonic geometry, numerical simulation analysis methods based on stress or strain fields, and multi-parameter weighted comprehensive evaluation methods.

[0003] Identification methods based on tectonic geometry typically classify structures according to fold amplitude, fault density, fault combination patterns, and tectonic zoning characteristics. While intuitive and easy to use, these methods heavily rely on the experience of interpreters, have highly subjective boundary delineation, and struggle to establish unified quantitative criteria. Numerical simulation analysis methods based on stress or strain fields can reproduce tectonic evolution processes and output response parameters such as displacement, strain, stress, and fracture activity. Discrete element method (DEM) simulations have been used to study the deformation of fold-thrust zones, foreland basins, and multi-detachment structures, but most results remain at the level of simulation interpretation, lacking standardized evaluation indicators for unified comparison across parameter combinations. Multi-parameter weighted comprehensive evaluation methods can superimpose factors such as fault density, tectonic amplitude, detachment layer thickness, and seismic properties for tectonic deformation intensity zoning; however, parameter weights often rely on manual experience, resulting in weak comparability between different study areas.

[0004] This invention aims to solve the following technical problems in existing methods for identifying structural deformation zones: First, the identification results are highly dependent on the experience of the interpreters, and the boundary delineation is subjective; second, the discrete element simulation results under different conditions of total slip layer thickness H, boundary loading rate v, and slip layer friction coefficient lack a unified quantitative comparison standard; third, although the discrete element simulation results can reflect the deformation concentration area, there is a lack of standardized evaluation coefficients and threshold determination procedures that can be directly used for structural deformation intensity classification.

[0005] To address the aforementioned problems, this invention provides a quantitative evaluation method for structural deformation intensity based on discrete element method and neural network. This method includes: parameter combination design, discrete element simulation and particle displacement sample construction, neural network displacement response prediction, quantitative Q-value characterization, K-value normalization and grading, and mapping to real seismic profiles. Summary of the Invention

[0006] To address the problems of existing methods for identifying tectonic deformation zones relying on human experience, lacking a unified comparison standard for discrete element simulation results under different external conditions, and lacking standardized evaluation coefficients and grading procedures, this invention provides a quantitative evaluation method for tectonic deformation intensity based on discrete element method and neural network.

[0007] To achieve the above objectives, the present invention includes the following steps:

[0008] S1. Based on the number of slip layers in the area to be evaluated and the combination of the average thickness, minimum thickness and maximum thickness of each slip layer, determine the total thickness H of the slip layer, and combine the boundary loading rate v and the slip layer friction coefficient μ to establish the combination of discrete element numerical simulation experimental parameters using the control variable method.

[0009] The boundary loading rate v and friction coefficient μ are set as follows: the boundary loading rate v is set to 1.0 m / s, 1.5 m / s and 2.0 m / s, respectively, as the slow, medium and fast tectonic evolution rates; the friction coefficient of the non-slip layer is uniformly set to 0.3 as the background mechanical parameter; the friction coefficient μ of the slip layer is set to 0 and 0.1, where μ=0 represents a strong slip layer and μ=0.1 represents a weak slip layer.

[0010] S2. Establish a two-dimensional discrete element cross-sectional model according to the parameter combination and conduct numerical simulation to derive the unique particle number, initial coordinates, and final state coordinates; match the initial and final states of the particles according to their particle numbers and calculate the vertical displacement ΔY. ij The model is constructed by including model number, particle coordinates, total thickness H of the slip layer, boundary loading rate v, slip layer friction coefficient μ, and vertical displacement ΔY of the particles. ij The particle training sample set.

[0011] S3. Constructing Input Features Z ij And with y ij To supervise the training of labels, a neural network regression model is trained. After training convergence, a fixed set of background reference samples is extracted from the model, and DeepExplainer is used to calculate the attribution value φH of the total slip layer thickness. ij Boundary loading rate φv ij and the attribution value of the friction coefficient of the slip layer φμ ij ;

[0012] The relevant formulas for the neural network regression model are:

[0013] fNN(Z ij )≈y ij =ln(ΔY ij +c)

[0014] In the formula, Z ij For input features; y ijThe labels are used to supervise training; c is a fixed numerical adjustment constant used to prevent logarithmic transformation from being impossible when the vertical displacement is equal to zero. Its value is selected according to the displacement resolution and remains globally unique throughout the entire evaluation task.

[0015] S4, with the vertical displacement of the particle ΔY ij Based on this, the SHAP local attribution values ​​of the total thickness of the detachment layer, the boundary loading rate, and the friction coefficient of the detachment layer are corrected, and the quantitative characterization coefficient Q of the equivalent structural deformation strength is calculated. ij .

[0016] S5. Using the minimum value Q of the training sample Q-value set. min and maximum value Q max As a fixed normalized benchmark for the same evaluation task, the structural deformation strength grading coefficient K is calculated according to the formula. ij And according to K ij The deformation strength of the structural zone is classified.

[0017] S6. Extract the X-coordinates of the evaluation point profile from the actual seismic profile. real and Y real Vertical displacement ΔY real Total thickness H of the slip layer real Boundary loading rate v real and the coefficient of friction of the slip layer μ real , and its Z real Input the neural network model solidified in step S3 and calculate the local attribution value of the boundary parameter SHAP; then calculate the quantitative characterization coefficient Q of the true equivalent structural deformation strength in steps S4 and S5 respectively. real The actual structural deformation strength grading coefficient K real The deformation intensity of the structural zone is divided for the evaluation points or evaluation segments of the actual seismic profile.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] (1) This invention establishes a quantitative evaluation process that combines structural deformation feature extraction, parameter correction and intensity grading, so as to identify structural deformation zones with unified evaluation indicators, reduce the reliance on the experience of interpreters and the subjectivity of boundary division, and improve the objectivity and repeatability of evaluation results.

[0020] (2) This invention utilizes neural network learning to construct deformation response relationships and corrects the influence of total slip layer thickness, boundary loading rate and friction coefficient through SHAP local attribution, providing a unified and comparable quantitative evaluation standard for discrete element simulation results under different parameter combinations;

[0021] (3) The present invention calculates the grading coefficient K based on the equivalent structural deformation strength Q value and the fixed normalization benchmark, and divides the weak, medium and strong structural deformation zones according to the unified threshold, forming a standardized process for evaluation coefficient calculation, normalization processing and grade determination. Attached Figure Description

[0022] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0023] Figure 1 This is a flowchart of a method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to the present invention;

[0024] Figure 2 This is a diagram of the initial model for discrete element numerical simulation according to an embodiment of the present invention.

[0025] Figure 3 This is a diagram of the final model of the discrete element numerical simulation according to an embodiment of the present invention.

[0026] Figure 4 This is a diagram of the neural network model architecture according to an embodiment of the present invention;

[0027] Figure 5 This is a comparison of the structural intensity distribution curves of Model 1 in this embodiment of the invention before and after SHAP local attribution correction;

[0028] Figure 6 This is a diagram showing the results of dividing a real seismic profile into strong tectonic deformation zones, moderate tectonic deformation zones, and weak tectonic deformation zones according to an embodiment of the present invention. Detailed Implementation

[0029] The following uses the eastern Sichuan region as an example to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention, such as... Figure 1 As shown, the specific solution of the present invention is described below:

[0030] S1. Based on the geological structure conditions of the area to be evaluated (thickness h of a single detachment layer, total thickness H of detachment layers), the thickness and number of detachment layers are set using the principle of similarity. Detachment layer thickness h setting scheme: There are n detachment layers in the target area, numbered T1, T2, T3…T from bottom to top. n Based on the average thickness h of each slip layer a Minimum thickness h min and maximum thickness h maxDetermine the thickness value, and sum the thicknesses of each slip layer to obtain the total slip layer thickness H under this model. When designing the slip layer thickness, h... a h min and h max The number of thickness values ​​for each type is kept basically consistent to ensure that the total thickness of the slip layer in the discrete element experimental group covers the actual range of the target area. The boundary loading rate v and friction coefficient μ are set as follows: In the discrete element structural numerical simulation, the boundary loading rate v is set to 1.0 m / s, 1.5 m / s, and 2.0 m / s, representing slow, medium, and fast tectonic evolution rates, respectively, which can represent most tectonic backgrounds. The friction coefficient of the non-slip layer is uniformly set to 0.3 as a background mechanical parameter. The friction coefficient μ of the slip layer is set to 0 and 0.1, where μ=0 represents a strong slip layer and μ=0.1 represents a weak slip layer. Specific settings are shown in Table 1.

[0031] Table 1. Range of Discrete Element Experiment Parameters

[0032]

[0033] Three detachment layers exist in the eastern Sichuan region, namely T1, T2, and T3 from bottom to top. The average thickness of the T1 detachment layer is 650 m, the minimum thickness is 300 m, and the maximum thickness is 1000 m; the average thickness of the T2 detachment layer is 900 m, the minimum thickness is 600 m, and the maximum thickness is 1200 m; and the average thickness of the T3 detachment layer is 500 m, the minimum thickness is 300 m, and the maximum thickness is 700 m. The specific settings of the discrete element method numerical simulation parameters are shown in Table 2.

[0034] Table 2. Parameter values ​​for three sets of detachment layers and discrete element simulations in eastern Sichuan.

[0035]

[0036] S2. Numerical Simulation Experiment and Data Export: Based on the combination of parameters determined in step S1, including the slip layer thickness, boundary loading rate, and slip layer friction coefficient, discrete element numerical simulation was conducted. The initial and final coordinates of the particles were exported and matched. A particle sample set containing model number, boundary parameters, particle profile coordinates, and vertical displacement was constructed, as detailed below:

[0037] ① Combining the target block slip layer thickness h, slip layer friction coefficient μ, and boundary loading rate v, M sets of discrete element experiments are designed using the controlled variable method. If the number of slip layers in the study area is 1, M = 1 × 3 × 2 = 6; if the number of slip layers in the study area is 2, M = 3 × 3 × 2 = 18; if the number of slip layers in the study area is 3, M = 6 × 3 × 2 = 36; if the number of slip layers in the study area is 4, M = 10 × 3 × 2 = 60. Reducing the number of sets M can reduce the computational load, but it will also reduce the parameter space coverage, model training, and stability of SHAP local attribution values. Increasing the number of sets M can improve the response surface coverage, but it will increase the computational cost. In specific implementation, the number of sets can be increased or decreased according to the actual situation.

[0038] ② Use Vbox discrete element modeling software to establish a two-dimensional discrete element construction profile model, and then perform M sets of discrete element numerical simulations to obtain the structural deformation results under different combinations of H, v, and μ. After each set of discrete element simulations is completed, export the unique particle number, initial particle coordinates, and final particle coordinates; organize the exported particle coordinate data to obtain a particle point dataset; match the initial and final coordinates of the same particle according to the unique particle number, and read the initial and final Y coordinates of the particle to calculate the vertical displacement ΔY of the particle. ij :

[0039]

[0040] In the formula, i is the particle sample number within the same model, j is the discrete element experimental model number, and Y... i1 Let Y be the initial Y coordinate of the particle. i2 Let Y be the final Y-coordinate of the particle, ΔY ij This represents the vertical displacement of the particle.

[0041] In the eastern Sichuan region, six thickness combinations were used, with three levels of boundary loading rate and two levels of slip layer friction coefficient. Therefore, the number of experimental groups M = 6 × 3 × 2 = 36. 36 discrete element numerical simulations were performed using Vbox discrete element modeling software (Table 3), generating a discrete particle set within the model region. A two-dimensional discrete element construction profile model was established. Taking Model 1 as an example, the initial model design is shown in [Table 3]. Figure 2 After the operation is completed, the output includes the unique particle number, initial particle coordinates, and final particle coordinates; the structural deformation results under different combinations of H, v, and μ are obtained. Figure 3 ), and serve as the data source for neural network training samples and subsequent Q-value calculations.

[0042] As shown in Table 3, a two-dimensional discrete element profile model is established based on the listed parameters and submitted for calculation. After the simulation calculation is completed, a data file containing attributes such as unique particle number, spatial coordinates, and radius is exported. The profile position coordinates and corresponding numbers of the required particle point dataset are extracted from the data and organized into initial and final state coordinate tables. The data is then processed according to ΔY. ij =|Y i2 -Y i1 |Calculate the vertical displacement ΔY of the particles in each model group. ij (Table 4) Table 4 only lists some representative particles in Model 1. In actual calculations, all particles in each model group are processed.

[0043] Table 3. Discrete Element Experiment Designs for 36 Groups in Eastern Sichuan

[0044]

[0045] Table 4. Representative vertical displacement ΔY of particles ij Calculation example (H=2000m, v=2.0, μ=0)

[0046]

[0047] S3. A neural network is used to learn the parameter response relationship and analyze the local attribution values ​​of each parameter using the granular profile coordinates and discrete element model parameters as inputs, and the logarithm of the granular vertical displacement as the supervised training label. The neural network regression model is trained, and the local attribution values ​​of each input parameter are calculated based on a fixed background reference sample set. The details are as follows:

[0048] ① Read the total thickness H of the slip layer corresponding to model number j. j Boundary loading rate v j and the coefficient of friction μ of the slip layer j And the initial particle profile coordinates X obtained in step S2 i1 Y i1 and particle vertical displacement ΔY ij Constructing input features Z ij =[X i1 ,Y i1 H j / H0,v j / v0,μ j ] and supervised training labels y ij =ln(ΔY ij +c); where H0 and v0 are the median total thickness of the slip layer and the median boundary loading rate, respectively, and remain unchanged in the same evaluation task; c is a fixed numerical adjustment constant used to prevent the inability to perform logarithmic transformation when the vertical displacement is equal to zero. It is selected according to the displacement resolution and remains globally unique in the entire evaluation task.

[0049] ②With Z ij As input features, with y ij To supervise the training of labels, a feedforward multilayer perceptron (MLP) regression neural network model was built. The model architecture is shown in the appendix. Figure 4 The neural network consists of 5 input nodes, 16 neurons in the first hidden layer, 8 neurons in the second hidden layer, and 1 linear output node; the 5 input nodes receive the initial spatial coordinates X of each particle. i1 Y i1 Normalized total slip layer thickness H j / H0, Normalized boundary loading rate v j / v0 and the friction coefficient of the slip layer μ j The nonlinear mapping relationship between particle profile coordinates, model parameters, and particle vertical displacement is learned to obtain a convergent neural network model fNN(Z). ij )≈y ij =ln(ΔY ij +c); Using the mean squared error of samples (MSE) as the convergence verification metric, continue training until the model converges;

[0050] ③ Select 100 samples to form a fixed background reference sample set. Use DeepExplainer to calculate the SHAP local attribution value, and use the average output of the neural network model corresponding to this background reference sample set as the attribution benchmark φ0. The SHAP algorithm uses the difference between the predicted output and the predicted mean φ0 of the background reference sample set to distribute the difference through backpropagation within the neural network, and calculates the specific SHAP local attribution value of each auxiliary parameter. The neural network prediction result can be expressed as:

[0051] fNN(Z ij )=φ0+φX i1 +φY i1 +φH ij +φv ij +φμ ij ≈ln(ΔY ij +c)

[0052] In the formula, φX i1 and φY i1 These are the SHAP local attribution values ​​for the spatial location corresponding to the particle profile coordinates, φH. ij ,φv ij and φμ ij These are the SHAP local attribution values ​​corresponding to the total thickness of the slip layer, the boundary loading rate, and the slip layer friction coefficient, respectively; where the SHAP local attribution values ​​corresponding to the three parameters are φH. ij ,φv ij and φμ ijThe calculation result serves as the quantitative characterization coefficient Q for the equivalent structural deformation strength in subsequent calculations. ij The auxiliary correction amount.

[0053] Using 36 discrete element experiments in eastern Sichuan as samples, with H0=2000m, v0=1.5m / s, and c=1.0m, taking sample 1 of model 1 as an example, its input features are Z1=[1054.2,6968.7,1.00,1.33,0], and the original vertical displacement ΔY 11 =5.4m, corresponding to supervised training label y 11 =ln(5.4+1.0)=1.8563, the input features of some neural network training samples, the original particle vertical displacement and the corresponding supervised training labels are shown in Table 5; the logarithmic prediction value of the particle vertical displacement output by the linear output node is fNN(Z1)≈ln(ΔY). 11 +1)=1.8563; the calculated attribution benchmark φ0=6.39; the local attribution value φH of the total slip layer thickness SHAP. ij =-0.3353, Boundary Loading Rate SHAP Local Attribution Value φv ij =-0.0033, the local attribution value of the slip layer friction coefficient SHAP φμ ij =0.1860.

[0054] Table 5. Comparison of Input Features of Neural Network Training Samples and Vertical Displacement of Original Particles

[0055]

[0056] S4. Construct a quantitative characterization coefficient Q for structural deformation strength based on SHAP local attribution value standardization correction. ij Based on the vertical displacement of particles, the SHAP local attribution values ​​corresponding to the total thickness of the slip layer, the boundary loading rate, and the friction coefficient of the slip layer are corrected on a logarithmic scale. This yields the equivalent structural deformation strength quantitative characterization coefficient Q, which corrects for the influence of external boundary conditions while preserving the spatial deformation characteristics. ij .

[0057] Determine the vertical displacement ΔY of the particle ij As the main control index for structural deformation strength, and with the total thickness of the detachment layer, the boundary loading rate, and the friction coefficient of the detachment layer as auxiliary correction parameters to correct deviations caused by different external conditions, it is necessary to correct the local attribution values ​​of these three boundary parameters from the predicted values ​​on a logarithmic scale to obtain the logarithmic space corrected displacement:

[0058] ln(Q ij +c)=ln(ΔY ij +c)-(φH ij +φvij +φμ ij )

[0059] By taking the exponents at both ends and rearranging the terms, we obtain the quantitative characterization coefficient Q of the equivalent structural deformation strength. ij :

[0060]

[0061] Corrected total slip layer thickness SHAP local attribution value φH ij Boundary loading rate SHAP local attribution value φv ij and the local attribution value φμ of the friction coefficient SHAP ij Preserve the local attribution value φX of the spatial location SHAP corresponding to the spatial location coordinates of the profile. ij and φY ij While correcting for external boundary condition interference, it retains the positional characteristics that reflect the spatial distribution of deformation intensity within the tectonic zone.

[0062] The calculated φH ij =-0.3353、φv ij =-0.0033、φμ ij =0.1860 Substitute into Q ij The formula is modified by nonlinear standardization to obtain the equivalent structural deformation strength characterization coefficient Q for sample 1 in eastern Sichuan. ij The value is 6.4551. The same calculation steps are performed on all discrete element experimental samples to obtain the Q value for the entire sample. ij value.

[0063] To visually illustrate the effect of the SHAP local attribution value correction on the spatial distribution characteristics of structural deformation, taking Model 1 as an example, the discrete element model is divided into 50 evaluation intervals at equal intervals along the horizontal direction of the profile, and the intensity distribution curves along the profile before and after correction are plotted, as shown below. Figure 5 As shown, the original vertical displacement ΔY before the correction of Model 1 ij The corrected equivalent structural deformation strength quantitative characterization coefficient Q ij The presence of largely consistent peak and valley locations indicates that the spatial distribution characteristics of structural deformation intensity within the model are preserved after correcting the local attribution values ​​corresponding to the total thickness of the slip layer, boundary loading rate, and slip layer friction coefficient; simultaneously, Q ij Curve relative to the original ΔY ij The adjustment in the numerical amplitude of the curve indicates that the local attribution offset corresponding to the aforementioned boundary parameters has been corrected.

[0064] S5. Normalize and determine the structural deformation intensity classification: Use the S4 training sample Q ijThe minimum and maximum values ​​of the value set are used as fixed normalization benchmarks for the same evaluation task to calculate the structural deformation strength grading coefficients and divide the structural deformation zones into weak, medium and strong.

[0065] ①Q obtained from S4 ij In the set of values, the minimum value Q is selected. min and maximum value Q max As a fixed normalization benchmark, calculate the structural deformation strength grading coefficient K. ij :

[0066]

[0067] ②Based on the structural deformation strength grading coefficient K ij The deformation strength of tectonic zones is classified into grades. The classification scheme for deformation strength of tectonic zones is as follows: when 0 ≤ K ij When <1 / 3, it is a weak structural zone; when 1 / 3≤K ij When <2 / 3, it is a medium-sized tectonic zone; when 2 / 3≤K ij When ≤1, it is a strong tectonic zone.

[0068] From the eastern Sichuan region Q ij Determine the minimum value Q from the set of values. min =0.8753, maximum value Q max =2928.9348, and these two values ​​are used as the fixed normalized benchmarks for the entire process of seismic profile evaluation in eastern Sichuan. Taking Sample 1 in Model 1 as an example, the equivalent structural deformation intensity quantitative characterization coefficient of 6.4551 is obtained by combining S4 calculation, and the structural deformation intensity grading coefficient K of Sample 1 is calculated. ij The value is 0.0019, therefore sample 1 is classified as a weakly structurally deformed sample, and... Figure 3 The deformation results of Sample 1 are consistent; the deformation strength grading coefficients of the remaining training samples are calculated using the same method, and their deformation strength levels are determined according to the specified grading thresholds.

[0069] S6. Evaluation of Structural Deformation Strength of Real Seismic Profiles: When evaluating the structural deformation strength of real seismic profiles, parameter extraction, parameter attribution correction, and strength normalization classification are performed sequentially, specifically including the following two steps:

[0070] ① For the actual seismic profile of the target work area, the actual vertical displacement ΔY is determined by the seismic interpretation horizon and the tectonic reconstruction reference surface. real And the total thickness H of the actual slip layer real Determine the true boundary loading rate v based on the structural deformation background. real Determine the friction coefficient μ of the true detachment layer based on the lithology of the detachment layer. real And read the true profile coordinates X of the evaluation point. real and Yreal The corresponding parameters are dimensionless using H0 and v0, which are determined and remain unchanged in step S3, to construct the input features Z of the real profile evaluation points. real =[X real ,Y real H real / H0,v real / v0,μ real ];

[0071] ②Z real Input the neural network model solidified in step S3, and call the same fixed background reference sample set to calculate the attribution values ​​of the total thickness of the true detachment layer, the true boundary loading rate, and the true detachment layer friction coefficient φH in the target work area. real ,φv real and φμ real Let Φ real =φH real +φv real +φμ real Press Q real =(ΔY real +c)×exp(-Φ real )-c Calculate the equivalent structural deformation strength quantitative characterization coefficient Q at the evaluation point of the actual profile. real Then, Q is fixed in step S5. min Q max The actual structural deformation strength grading coefficient K is calculated using the following formula. real The structural deformation zones are divided into weak, medium, and strong structural deformation zones according to the structural deformation strength classification scheme in step S5.

[0072]

[0073] Stratigraphic interpretation and tectonic reconstruction of seismic profiles in eastern Sichuan were carried out to achieve the following: Figure 6 The X coordinates of the read profile position are shown. real And the vertical coordinates of the target layer where the evaluation point is located are used as Y. real The absolute value of the vertical distance between the target layer at the evaluation point and the tectonic restoration reference surface is taken as the true vertical displacement ΔY. real Based on the seismic interpretation results, the true total thickness H of the detachment layer corresponding to each evaluation point was determined. real The true boundary loading rate v was determined based on the tectonic deformation background of the eastern Sichuan fold belt. real The friction coefficient μ of the actual detachment layer was determined based on the lithology of the detachment layer in eastern Sichuan. real .

[0074] The true total thickness H of the detachment layer at each evaluation point real Take 2000m as the actual boundary loading rate v realTake 2.0 m / s as the actual friction coefficient μ of the slip layer. real Taking 0, and performing dimensionless processing with H0=2000m and v0=1.5m / s, we obtain H. real / H0=1.00、v real / v0=1.33; Therefore, each real profile evaluation point is constructed as the neural network input feature Z. real =[X real ,Y real [,1.00,1.33,0], where Z corresponds to each evaluation point. real Input the neural network model fNN trained and solidified in step S3, load the same set of 100 fixed background reference samples determined in step S3, with an attribution benchmark of φ0=6.39, call the DeepExplainer calculation module, and according to the reverse chain allocation rule of the neural network structure, distribute the difference between the model's predicted output and the mean of the background prediction to the five input features, so that it satisfies fNN(Z real )-φ0=φX real +φY real +φH real +φv real +φμ real Thus, φH is obtained. real +φv real +φμ real Let Φ be a value. real =φH real +φv real +φμ real ΔY of each evaluation point real and Φ real Substituting into the following formula, calculate the quantitative characterization coefficient Q of the true equivalent structural deformation strength at the actual profile evaluation point. real :

[0075]

[0076] The fixed normalized reference Q determined in step S5 is used. min =0.8753 and Q max =2928.9348, calculate the actual structural deformation strength grading coefficient K. real The grading results of structural deformation intensity of the actual seismic profile in eastern Sichuan were obtained (Table 6). The grading results of evaluation points 1-8 were then applied along... Figure 6 The cross-sections are shown from left to right. Strong tectonic zones correspond to large vertical displacement and concentrated tectonic deformation features (e.g., evaluation point 1 in Table 6); moderate tectonic zones correspond to moderate vertical displacement and local bending deformation features of strata (e.g., evaluation point 2 in Table 6); and weak tectonic zones correspond to small vertical displacement and continuous strata features (e.g., evaluation points 3, 4, 5, 6, 7, and 8 in Table 6).

[0077] Table 6. Results of parameter extraction and structural deformation intensity classification for representative evaluation points on real seismic profiles.

[0078]

[0079] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the protection scope of the claims and specification of the present invention.

Claims

1. A quantitative evaluation method for structural deformation strength based on discrete element method and neural network, characterized in that, Includes the following steps: S1. Based on the number of slip layers in the area to be evaluated and the combination of the average thickness, minimum thickness and maximum thickness of each slip layer, determine the total thickness H of the slip layer, and combine the boundary loading rate v and the slip layer friction coefficient μ to establish the combination of discrete element numerical simulation experimental parameters using the control variable method. S2. Establish a two-dimensional discrete element cross-sectional model according to the parameter combination and conduct numerical simulation to derive the unique particle number, initial coordinates, and final state coordinates; match the initial and final states of the particles according to their particle numbers and calculate the vertical displacement ΔY. ij The model is constructed by including model number, particle coordinates, total thickness H of the slip layer, boundary loading rate v, slip layer friction coefficient μ, and vertical displacement ΔY of the particles. ij The particle training sample set; S3. Constructing Input Features Z ij And with y ij To supervise the training of labels, a neural network regression model is trained. After training convergence, a fixed set of background reference samples is extracted from the model, and DeepExplainer is used to calculate the attribution value φH of the total slip layer thickness. ij Boundary loading rate φv ij and the attribution value of the friction coefficient of the slip layer φμ ij ; S4, with the vertical displacement of the particle ΔY ij Based on this, the SHAP local attribution values ​​of the total thickness of the detachment layer, the boundary loading rate, and the friction coefficient of the detachment layer are corrected, and the quantitative characterization coefficient Q of the equivalent structural deformation strength is calculated. ij ; S5. Using the minimum value Q of the training sample Q-value set. min and maximum value Q max As a fixed normalized benchmark for the same evaluation task, the structural deformation strength grading coefficient K is calculated according to the formula. ij And according to K ij The deformation strength of the structural zone is classified. S6. Extract the X-coordinates of the evaluation point profile from the actual seismic profile. real and Y real Vertical displacement ΔY real Total thickness H of the slip layer real Boundary loading rate v real and the coefficient of friction of the slip layer μ real , and its Z real Input the neural network model solidified in step S3 and calculate the local attribution value of the boundary parameter SHAP; then calculate the quantitative characterization coefficient Q of the true equivalent structural deformation strength in steps S4 and S5 respectively. real The actual structural deformation strength grading coefficient K real The deformation intensity of the structural zone is divided for the evaluation points or evaluation segments of the actual seismic profile.

2. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, The boundary loading rate v and friction coefficient μ described in S1 are set as follows: the boundary loading rate v is set to 1.0 m / s, 1.5 m / s and 2.0 m / s, respectively, as slow, medium and fast tectonic evolution rates; the friction coefficient of the non-slip layer is uniformly set to 0.3 as a background mechanical parameter; the friction coefficient μ of the slip layer is set to 0 and 0.1, where μ=0 represents a strong slip layer and μ=0.1 represents a weak slip layer.

3. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, The vertical displacement ΔY of the particles mentioned in S2 ij The calculation formula is: In the formula, i is the particle sample number within the same model, j is the discrete element experimental model number, and Y... i1 Let Y be the initial Y coordinate of the particle. i2 Let Y be the final Y-coordinate of the particle, ΔY ij This represents the vertical displacement of the particle.

4. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, The relevant formulas for the neural network regression model described in S3 are: fNN(Z ij )≈y ij =ln(ΔY ij +c) In the formula, Z ij For input features; y ij The labels are used to supervise training; c is a fixed numerical adjustment constant used to prevent logarithmic transformation from being impossible when the vertical displacement is equal to zero. Its value is selected according to the displacement resolution and remains globally unique throughout the entire evaluation task.

5. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, The equivalent structural deformation strength quantitative characterization coefficient Q in S4 ij The calculation formula is: 。 6. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, The structural deformation strength grading coefficient K mentioned in S5 ij The calculation formula is: In the formula, Q min For Q ij The minimum value in the set of values; Q max For Q ij The maximum value in the set of values.

7. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, According to K as described in S5 ij The specific standard for classifying the deformation strength of tectonic zones is as follows: when 0 ≤ K ij When <1 / 3, it is a weak structural zone; when 1 / 3≤K ij When <2 / 3, it is a medium-sized tectonic zone; when 2 / 3≤K ij When ≤1, it is a strong tectonic zone.

8. The method for quantitatively evaluating the structural deformation strength based on discrete element method and neural network according to claim 1, characterized in that, The quantitative characterization coefficient Q of the true equivalent structural deformation strength described in S6 real The calculation formula is: Q real =(ΔY real +c)×exp(-Φ real )-c In the formula, ΔY real The actual vertical displacement ΔY of the target work area real ;Φ real =φH real +φv real +φμ real ,φH real ,φv real and φμ real The attribution values ​​are the actual total thickness of the slip layer, the actual boundary loading rate, and the actual slip layer friction coefficient for the target work area. The actual structural deformation strength grading coefficient K real The calculation formula is: 。

Images