A method for evaluating shear strength of rock mass structural planes based on neural network

Abstract

The present invention provides a method for assessing the shear strength of rock structural surfaces based on a neural network, and relates to the technical field of rock shear strength assessment. The present invention improves the accuracy of shear strength assessment of rock structural surfaces by integrating three-dimensional morphology analysis with multi-scale neural network modeling. The directional weight matrix quantifies the influence of directional heterogeneity of the structural surface on the shear strength, and combines the surface undulation parameters with the equivalent compressive strength formula to avoid the problem of insufficient characterization of multi-scale morphology. Secondly, a dynamic correction factor and an adaptive weight optimization algorithm are introduced to effectively compensate for size effects and material sensitivity differences, and avoid the model from falling into local optimality. The multi-scale feature fusion mechanism simultaneously extracts macroscopic morphological laws and microscopic undulation details, and generates strength grade labels through dynamic grading functions, which simplifies the complexity of engineering decision-making. Finally, the three-dimensional visual spatial mapping intuitively presents the relationship between structural surface morphology and strength distribution, enhancing the interpretability of the results.

Description

Technical Field

[0001] The present invention relates to the technical field of rock mass shear strength assessment, and in particular to a rock mass structural surface shear strength assessment method based on a neural network. Background Art

[0002] Assessing the shear strength of rock mass structural surfaces is a core component of geotechnical engineering stability analysis. Traditional methods rely on empirical formulas or single mechanical parameters (such as normal stress and internal friction angle) for estimation, which presents significant limitations. Existing techniques often use the harmonic mean method to calculate the equivalent strength of heterogeneous rock masses. However, this method ignores the nonlinear effects of the strength difference between the upper and lower walls on shear behavior, leading to a systematic underestimation of the strength of asymmetric rock masses. Furthermore, the three-dimensional morphological characteristics of rock mass structural surfaces (such as directional distribution and surface undulation) are not fully quantified, and are instead characterized using simplified parameters such as the roughness coefficient, making it difficult to reflect the coupled effects of multi-scale morphology on shear strength.

[0003] In the prior art, publication number CN105466790A discloses a method for evaluating the shear strength of anisotropic rock structural surfaces. The method includes the following steps: (1) obtaining three-dimensional morphological data of the rock structural surface; (2) calculating the structural surface anisotropic characteristic parameter SRv and the fluctuation amplitude parameter A, and using the two parameters SRv and A to represent the roughness of the three-dimensional rock structural surface in all directions; (3) testing the wall strength JCS and basic friction angle of the same type of rock structural surface; (4) substituting the structural surface parameters obtained in steps (2) and (3) into the formula to obtain the peak shear stress under different normal stresses. It also relies on empirical formulas and limited morphological parameters, cannot fully characterize multi-scale morphological characteristics, and does not introduce dynamic correction and adaptive optimization mechanisms, resulting in insufficient adaptability to non-uniform rock masses, size effects, and complex mechanical responses, and limited prediction accuracy.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a rock mass structural surface shear strength evaluation method based on neural network to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for evaluating the shear strength of rock mass structural surfaces based on a neural network comprises the following steps:

[0008] S1: Collect 3D topographic data and mechanical data of rock mass structural surface, extract topographic characteristic parameters and generate correction factors respectively;

[0009] S2: Construct a multi-scale neural network model, input morphological feature parameters, mechanical data and correction factors, and perform training optimization through an adaptive weight optimization algorithm;

[0010] S3: Use the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural surface, and generate the corresponding strength grade label based on the statistical distribution characteristics;

[0011] S4: Correlate the 3D topography data, predicted shear strength values, and strength grade labels to generate a visual spatial map.

[0012] Preferably, the topographic characteristic parameters include directional distribution characteristics and surface relief parameters;

[0013] The calculation method of the directional distribution characteristics is:

[0014] ;

[0015] In the formula Indicates the The direction angle in each direction, Indicates the The directional weight matrix of the direction angle in the direction, Indicates the The dominant direction angle in each direction, Indicates the The weight coefficients in each direction, ,and , subscript Indicates direction index, superscript Also represents the direction index, Indicates the total number of directions, represents the attenuation coefficient, and .

[0016] Preferably, the surface relief parameters include apparent dip statistics and height distribution statistics;

[0017] When collecting the three-dimensional topographic data and mechanical data of the rock mass structural surface, several groups of sampling points are set, and the apparent dip statistics are the mean and root mean square of the apparent dip angles of the sampling points;

[0018] The height distribution statistics are the mean and root mean square of the relative heights of the sampling points.

[0019] Preferably, the mechanical data include normal stress, basic internal friction angle, uniaxial compressive strength and equivalent compressive strength of the rock mass structural surface and lower wall rock;

[0020] The equivalent compressive strength is calculated based on the uniaxial compressive strength of the rock mass structural surface and the lower wall rock. The calculation formula is:

[0021] ;

[0022] In the formula represents the equivalent compressive strength, 、 They represent the uniaxial compressive strength of the rock mass structural surface and the lower and upper wall rocks, represents the base coupling index, and , represents the difference correction factor, which is calculated as:

[0023] ;

[0024] In the formula represents the material sensitivity coefficient, and .

[0025] Preferably, the correction factor is calculated as follows:

[0026] ;

[0027] In the formula represents the correction factor, Indicates the length of the rock mass structural surface, Indicates the preset reference length, 、 represents the adjustment factor of the formula, , , represents the nonlinear compensation term, Indicates a sign function. When the value in the brackets is greater than 0, the function value is 1, and when the value in the brackets is less than or equal to 0, the function value is -1.

[0028] The calculation method of the nonlinear compensation term is:

[0029] ;

[0030] In the formula represents the dynamic adjustment factor, and .

[0031] Preferably, the adaptive weight optimization algorithm adopts a dynamic adaptive cuckoo algorithm, and its step size adjustment strategy is set as:

[0032] ;

[0033] In the formula Indicates the The step size for the iteration, Indicates the preset initial maximum step size, , represents the hyperbolic secant function, represents the decay rate factor, The index representing the number of iterations, represents the maximum number of iterations, and .

[0034] Preferably, the logic of using the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural surface is:

[0035] The morphological characteristic parameters, mechanical data and correction factors are input into the multi-scale neural network model to extract the macroscopic morphological characteristic vector and the microscopic relief characteristic vector respectively.

[0036] The morphological feature vector and the relief feature vector are processed through the multi-scale feature fusion layer to extract the fused feature vector, which is expressed as:

[0037] ;

[0038] In the formula 、 、 Represent the fusion feature vector, the morphology feature vector, and the undulation feature vector respectively, represents the weight matrix of the multi-scale feature fusion layer, represents the activation function, represents the feature concatenation operation, Represents the bias of the multi-scale feature fusion layer;

[0039] Then, the predicted value of shear strength is calculated by linear mapping in the output layer, and the expression is:

[0040] ;

[0041] In the formula represents the predicted value of shear strength, 、 denote the weight matrix and bias of the output layer respectively.

[0042] Preferably, the logic for generating corresponding intensity level labels in combination with statistical distribution characteristics is:

[0043] Collect three-dimensional morphological data and mechanical data of different rock mass structural surfaces, and calculate the corresponding predicted value of shear strength;

[0044] The predicted values ​​of shear strength of different rock mass structural planes are used as historical data, and their mean and standard deviation are calculated;

[0045] A dynamic grading function is constructed to output the strength grade label corresponding to the predicted value of shear strength. The dynamic grading function expression is:

[0046] ;

[0047] In the formula Indicates the grade strength label, Indicates the preset reference value of shear strength, represents the standard deviation of historical data, represents the volatility factor, , Represents the clipping function, constraint , Indicates rounding down within the brackets;

[0048] Grade Strength Label , and its confidence interval is marked as ,in , the predicted value of shear strength is proportional to the value of the grade strength label.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention significantly improves the accuracy and adaptability of shear strength assessment of rock structural surfaces by integrating three-dimensional morphology analysis and multi-scale neural network modeling. First, the directional weight matrix quantifies the influence of directional heterogeneity of the structural surface on the shear strength. Combined with the surface undulation parameters and the equivalent compressive strength formula, it solves the problem of insufficient characterization of non-uniform rock masses and multi-scale morphologies by traditional methods. Secondly, the dynamic correction factor and adaptive weight optimization algorithm are introduced to effectively compensate for the size effect and material sensitivity differences, and avoid the model from falling into local optimality. The multi-scale feature fusion mechanism simultaneously extracts macroscopic morphological laws and microscopic undulation details, and generates strength grade labels through dynamic grading functions, which simplifies the complexity of engineering decision-making. Finally, the three-dimensional visual spatial mapping intuitively presents the relationship between structural surface morphology and strength distribution, enhancing the interpretability of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0053] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0054] Example:

[0055] See also Figure 1 , the present invention provides a technical solution:

[0056] A method for evaluating the shear strength of rock mass structural surfaces based on a neural network comprises the following steps:

[0057] S1: Collect three-dimensional morphological data and mechanical data of the rock mass structural surface, extract morphological characteristic parameters and generate correction factors respectively.

[0058] The morphological characteristic parameters include directional distribution characteristics and surface relief parameters;

[0059] The calculation method of the directional distribution characteristics is:

[0060] ;

[0061] In the formula Indicates the The direction angle in each direction, Indicates the The directional weight matrix of the direction angle in the direction, Indicates the The dominant direction angle in each direction. The dominant direction angle refers to the dominant geometric direction distribution angle in the rock mass structural surface. It is usually consistent with the direction in which shear sliding is likely to occur in the structural surface. It is used to reflect the directional characteristics of the rock mass surface morphology. When setting it specifically, you can first calculate the normal vector of each local area and convert it into a direction angle. Then, you can count the distribution frequency of all local direction angles and identify the angle intervals that appear frequently. Finally, use algorithms such as K-means and DBSCAN to cluster the direction angle data and classify similar directions as the same dominant direction. Indicates the The weight coefficients in each direction, and The specific setting method can be selected according to the actual situation: for example, by measuring the frequency of occurrence of different direction angles in the rock mass structural surface, the frequency value is normalized and directly used as the weight coefficient; or the principal component analysis is performed on the structural surface direction data to extract the main direction (the direction with the largest variance contribution), and the weight is assigned according to the variance contribution rate of the principal component; or directly the expert experience is used to set the score, etc. Indicates direction index, superscript Also represents the direction index, Indicates the total number of directions, represents the attenuation coefficient, and .

[0062] The calculation principle of the directional distribution characteristics here is to calculate the contribution weight of different direction angles to the current analysis direction through the weight coefficient and attenuation coefficient. The exponential term in the formula is It is used to represent the influence of direction difference on weight, and can identify the dominant sliding direction of the rock mass structural surface, that is, the direction with lower shear strength.

[0063] When collecting the three-dimensional morphological data and mechanical data of the rock mass structural surface, several groups of sampling points are set. When setting the sampling points, the entire area of ​​the rock mass structural surface should be covered, including the area with significant undulation (such as bulges and depressions), the direction-dominant area (such as the dominant joint direction) and the transition area (the area with gradual morphological changes), avoiding local dense or sparse sampling to ensure uniform spatial distribution. The specific setting is determined according to the actual rock mass structural surface, which will not be elaborated here. The surface undulation parameters include the apparent dip statistics and the height distribution statistics;

[0064] The statistics of the apparent dip angle are the mean and root mean square of the apparent dip angles at the sampling points;

[0065] The height distribution statistics are the mean and root mean square of the relative heights of the sampling points.

[0066] Mechanical data include normal stress, basic internal friction angle, uniaxial compressive strength of the rock mass's lower and upper walls, and equivalent compressive strength. Normal stress and basic internal friction angle can be obtained by performing triaxial compression tests on samples from the sampling points in the laboratory. Normal stress can be obtained by applying normal loads under different confining pressures and inversely calculating the failure criterion (such as Mohr-Coulomb). Basic internal friction angle can be determined by applying confining pressure to the sample, gradually increasing the axial stress until failure, and then plotting the Mohr circle. The slope of the envelope can be used to determine the basic internal friction angle. The uniaxial compressive strength of the rock mass's lower and upper walls can be obtained through uniaxial compression testing: After processing samples collected from the upper and lower walls into standard specimens, an axial load is applied at a constant rate (e.g., 0.5-1 MPa / s). The maximum stress at failure is recorded, which is the uniaxial compressive strength.

[0067] The equivalent compressive strength is calculated based on the uniaxial compressive strength of the rock mass structural surface and the lower wall rock. The calculation formula is:

[0068] ;

[0069] In the formula represents the equivalent compressive strength, 、 They represent the uniaxial compressive strength of the rock mass structural surface and the lower and upper wall rocks, represents the base coupling index, and , represents the difference correction factor, which is calculated as:

[0070] ;

[0071] In the formula represents the material sensitivity coefficient, and .

[0072] From the calculation formula of equivalent compressive strength, we can see that its basic principle is based on the harmonic mean, and introduces a difference correction factor to reflect the influence of the strength difference between the upper and lower plates on the equivalent strength. The basic term in the brackets is used to reflect the equilibrium effect of the strength of the upper and lower plates of the rock mass structure, and the exponential term is used to amplify the difference effect. hour , improve the equivalent strength, and vice versa reduce it, thus avoiding the underestimation of asymmetric strength by traditional harmonic mean.

[0073] The correction factor is calculated as:

[0074] ;

[0075] In the formula represents the correction factor, Indicates the length of the rock mass structural surface, Indicates the preset reference length, 、 represents the adjustment factor of the formula, , , the values ​​of both can be fitted through historical data or determined based on expert experience. represents the nonlinear compensation term, Indicates a sign function. When the value in the brackets is greater than 0, the function value is 1, and when the value in the brackets is less than or equal to 0, the function value is -1.

[0076] The calculation method of the nonlinear compensation term is:

[0077] ;

[0078] In the formula represents the dynamic adjustment factor, and .

[0079] From the calculation formula of the correction factor, we can see that 、 They are used to control the response of the linear and quadratic terms to the length scale, thereby adapting to the size effect at different scales. It is used to realize dynamic adjustment compensation reverse, that is Time compensation , Time compensation , thereby reducing the nonlinear error caused by length mutation and improving the calculation accuracy of the model.

[0080] In this step, by collecting 3D topographic and mechanical data of the rock mass's structural surfaces, key characteristic parameters (morphological parameters and correction factors) are extracted, providing input for subsequent model training. Specifically, the 3D topographic data captures the macroscopic and microscopic geometric characteristics of the structural surfaces (such as directional distribution and surface undulation), compensating for the shortcomings of traditional single mechanical parameters. The correction factors, by introducing dynamic parameters such as the length of the structural surface, address the traditional model's neglect of scale effects and improve predictive adaptability. The equivalent compressive strength formula (JCS_K) of the mechanical data incorporates the difference in rock strength between the upper and lower walls, enhancing the characterization of heterogeneous rock masses.

[0081] S2: Construct a multi-scale neural network model, input morphological feature parameters, mechanical data and correction factors, and perform training optimization through an adaptive weight optimization algorithm;

[0082] In step S2, the adaptive weight optimization algorithm adopts the dynamic adaptive cuckoo algorithm, and its step size adjustment strategy is set as:

[0083] ;

[0084] In the formula Indicates the The step size for the iteration, Indicates the preset initial maximum step size, , represents the hyperbolic secant function, represents the decay rate factor, The index representing the number of iterations, represents the maximum number of iterations, and .

[0085] Specifically, the logic of the dynamic adaptation cuckoo algorithm is:

[0086] During the initialization phase, multiple sets of neural network weight combinations (each set is called a "bird's nest") are randomly created to cover different parameter configuration possibilities. Parameters such as the step size range and elimination probability are initialized to give the algorithm the basic ability to make adaptive adjustments. For example, morphological feature parameters, mechanical data, and correction factors are used as input parameters to randomly generate multiple sets of weight combinations, that is, multiple "bird's nests", which are used to map the input parameters to the predicted value of shear strength.

[0087] During the fitness evaluation phase, each weighted set is used to predict shear strength on the validation set. The error between the predicted and true values ​​is used as the evaluation criterion (the smaller the error, the better the weight). All weighted combinations are ranked by error, and the current optimal solution is retained. It is understood that the error here can be calculated using a variety of different methods, with mean square error being the most commonly used.

[0088] In the Levy search phase, non-optimal weight combinations are updated according to the random pattern of Levy flight, that is, long jumps avoid local optimality, short steps for fine search, and the step size is automatically adjusted according to the population diversity. Referring to the calculation formula in the above step size adjustment strategy, its decay rate factor is affected by the population diversity. In the early stage of iteration, the population diversity is large, the decay rate factor is set to a small value, and a large step size is used for global exploration, allowing individuals to be widely distributed in the search space. As the number of iterations increases, the population diversity slowly decreases, the decay rate factor is set to a larger value, the step size decays faster, and the algorithm focuses on fine search in local areas, while improving convergence efficiency. The decay rate factor can be set dynamically according to the number of iterations, for example, in When , it is fixed to 0.1 to force a global search. When set between 0.1 and 0.5, balance the relationship between search and development, and finally In the stage of , set it to 0.5 to accelerate convergence. and The relationship between them is set in proportion, which will not be explained here one by one;

[0089] The population's dispersion is then monitored, and the distribution differences among all weight combinations are calculated (larger differences indicate greater diversity). If the population is too concentrated, poorly performing weight combinations are probabilistically eliminated, and new solutions are generated near the optimal solution to maintain search activity. The search stops when the prediction error no longer decreases significantly (e.g., the change is less than a threshold for five consecutive iterations) or when the maximum number of iterations is reached.

[0090] Since the dynamic adaptive cuckoo algorithm is an existing technology, its specific calculation formula and calculation method are not described in detail here.

[0091] During model training, the measured shear strength values ​​of the samples collected in the above steps can be measured and then input into the model for training. Specifically, the measured shear strength values ​​can be obtained in the laboratory by applying shear force to rock surface samples. The shear force is gradually applied until the surface slips, and the maximum shear stress is recorded as the shear strength. The data from the rock surface sampling points is divided into an 80% training set and a 20% validation set. The mean squared error (MSE) is used to measure the difference between the predicted and measured shear strength values, and the model is then updated using a backpropagation algorithm. If the mean squared error remains below a preset convergence threshold over several consecutive model iterations, the model is considered optimized. The specific number of iterations can be determined based on expert experience. A maximum number of iterations can also be set to force convergence when the maximum number of iterations is reached, and the historical optimal weight (i.e., the weight corresponding to the minimum mean squared error) is used.

[0092] In this step, a multi-scale neural network model was constructed and trained using an adaptive weight optimization algorithm (dynamic adaptive cuckoo algorithm), achieving feature fusion and model optimization. Specifically, multi-scale feature extraction can simultaneously capture macroscopic morphology (such as directional distribution) and microscopic fluctuations (such as obliquity statistics), avoiding the limitations of single-scale features. Adaptive weight optimization can balance global search and local convergence by dynamically adjusting the step size, preventing the model from falling into local optimality. The use of correction factors as input parameters enhances the model's adaptability to actual engineering conditions.

[0093] S3: Use the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural surface, and generate the corresponding strength grade label based on the statistical distribution characteristics.

[0094] In step S3, the logic of using the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural surface is:

[0095] The morphological characteristic parameters, mechanical data and correction factors are input into the multi-scale neural network model to extract the macroscopic morphological characteristic vector and the microscopic relief characteristic vector respectively.

[0096] The morphological feature vector and the relief feature vector are processed through the multi-scale feature fusion layer to extract the fused feature vector, which is expressed as:

[0097] ;

[0098] In the formula 、 、 Represent the fusion feature vector, the morphology feature vector, and the undulation feature vector respectively, represents the weight matrix of the multi-scale feature fusion layer, represents the activation function, represents the feature concatenation operation, Represents the bias of the multi-scale feature fusion layer;

[0099] Then, the predicted value of shear strength is calculated by linear mapping in the output layer, and the expression is:

[0100] ;

[0101] In the formula represents the predicted value of shear strength, 、 denote the weight matrix and bias of the output layer respectively.

[0102] Specifically, macroscopic morphological features can include overall relief trends (such as the direction of the dominant dip angle and average height distribution), regional structural features (such as the dominant orientation of joint sets and the distribution of large asperities), and the spatial distribution pattern of the directional weight matrix (the polar coordinate statistical properties of the directional weight matrix). These are used to characterize the large-scale morphological regularities of rock mass structural surfaces and can be extracted using a 5x5 convolution kernel in a multiscale neural network. Microscopic relief feature vectors can include fluctuations in the standard deviation of the apparent dip angle, roughness details (such as micron-scale surface texture and crack density), and local outliers (i.e., regions of sudden changes) in the directional weight matrix. These are used to characterize local details of the structural surface and can be extracted using a 1x1 convolution kernel in a multiscale neural network. The integration of macroscopic and microscopic features is essentially a multi-scale information complementarity. Macroscopic features provide overall strength trends (such as the influence of the dominant dip angle), while microscopic features correct for local anomalies (such as strength attenuation caused by excessive roughness), thereby improving model accuracy and avoiding the limitations of traditional single-scale models.

[0103] In step S3, the logic for generating corresponding intensity level labels based on statistical distribution characteristics is as follows:

[0104] Collect three-dimensional morphological data and mechanical data of different rock mass structural surfaces, and calculate the corresponding predicted value of shear strength;

[0105] The predicted values ​​of shear strength of different rock mass structural planes are used as historical data, and their mean and standard deviation are calculated;

[0106] A dynamic grading function is constructed to output the strength grade label corresponding to the predicted value of shear strength. The dynamic grading function expression is:

[0107] ;

[0108] In the formula Indicates the grade strength label, It indicates the preset reference value of shear strength, the specific value of which can be determined based on expert experience or engineering standards. represents the standard deviation of historical data, represents the volatility factor, , Represents the clipping function, constraint , Indicates rounding down within the brackets;

[0109] Grade Strength Label , and its confidence interval is marked as ,in , the predicted value of shear strength is proportional to the value of the grade strength label.

[0110] Here, the predicted value of shear strength is standardized by the mean and standard deviation of historical data. Each level corresponds to a standard deviation interval, making it more intuitive and able to simplify engineering decision-making.

[0111] In this step, by splicing macro and micro features for multi-scale feature fusion, the predicted value of shear strength can be calculated more accurately. The mean and standard deviation of historical data are then used to construct classification rules, realize adaptive updating of classification results, and provide risk warnings for engineering decisions.

[0112] S4: Correlate the 3D topography data, predicted shear strength values, and strength grade labels to generate a visual spatial map.

[0113] In this step, the correspondence between the structural surface morphology and the strength grade is intuitively displayed through three-dimensional visualization, assisting engineers in quickly locating high-risk areas, enhancing the interpretability of the results, and facilitating comparison and verification with on-site measured data.

[0114] In summary, the present invention significantly improves the accuracy and adaptability of shear strength assessment of rock structural surfaces by integrating three-dimensional morphology analysis and multi-scale neural network modeling. First, the directional weight matrix quantifies the influence of directional heterogeneity of the structural surface on the shear strength. Combined with the surface undulation parameters and the equivalent compressive strength formula, it solves the problem of insufficient characterization of non-uniform rock masses and multi-scale morphologies by traditional methods. Secondly, the dynamic correction factor and the adaptive weight optimization algorithm are introduced to effectively compensate for the size effect and material sensitivity differences, and avoid the model from falling into local optimality. The multi-scale feature fusion mechanism simultaneously extracts macroscopic morphological laws and microscopic undulation details, and generates strength grade labels through dynamic grading functions, which simplifies the complexity of engineering decision-making. Finally, the three-dimensional visual spatial mapping intuitively presents the relationship between structural surface morphology and strength distribution, enhancing the interpretability of the results.

[0115] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0116] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0117] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0118] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for evaluating the shear strength of rock mass structural surfaces based on a neural network, characterized in that: The specific steps include: S1: Collect 3D topographic data and mechanical data of rock mass structural surface, and obtain topographic characteristic parameters and correction factors based on them; The topographic characteristic parameters include directional distribution characteristics and surface relief parameters; The calculation method of the directional distribution characteristics is: In the formula Indicates the The direction angle in each direction, Indicates the The directional weight matrix of the direction angle in the direction, Indicates the The dominant direction angle in each direction, Indicates the The weight coefficients in each direction, ,and , subscript Indicates direction index, superscript Also represents the direction index, Indicates the total number of directions, represents the attenuation coefficient, and ; The mechanical data include normal stress, basic internal friction angle, uniaxial compressive strength of the rock mass structural surface and the lower wall rock, and equivalent compressive strength; The correction factor is calculated as follows: In the formula represents the correction factor, Indicates the length of the rock mass structural surface, Indicates the preset reference length, 、 represents the adjustment factor of the formula, , , represents the nonlinear compensation term, Indicates a sign function. When the value in the brackets is greater than 0, the function value is 1, and when the value in the brackets is less than or equal to 0, the function value is -1. The calculation method of the nonlinear compensation term is: In the formula represents the dynamic adjustment factor, and ; S2: Construct a multi-scale neural network model, input morphological feature parameters, mechanical data and correction factors, and perform training optimization through an adaptive weight optimization algorithm; S3: Use the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural surface, and generate the corresponding strength grade label based on the statistical distribution characteristics; S4: Correlate the 3D topography data, predicted shear strength values, and strength grade labels to generate a visual spatial map.

2. The method for evaluating shear strength of rock mass structural surfaces based on a neural network according to claim 1, wherein: The surface relief parameters include apparent dip statistics and height distribution statistics; When collecting the three-dimensional topographic data and mechanical data of the rock mass structural surface, several groups of sampling points are set, and the apparent dip statistics are the mean and root mean square of the apparent dip angles of the sampling points; The height distribution statistics are the mean and root mean square of the relative heights of the sampling points.

3. The method for evaluating shear strength of rock mass structural surfaces based on a neural network according to claim 1, wherein: The equivalent compressive strength is calculated based on the uniaxial compressive strength of the rock mass structural surface and the lower wall rock, and the calculation formula is: In the formula represents the equivalent compressive strength, 、 They represent the uniaxial compressive strength of the rock mass structural surface and the lower and upper wall rocks, represents the base coupling index, and , represents the difference correction factor, which is calculated as: In the formula represents the material sensitivity coefficient, and .

4. The method for evaluating shear strength of rock mass structural surfaces based on a neural network according to claim 1, wherein: The adaptive weight optimization algorithm adopts the dynamic adaptive cuckoo algorithm, and its step size adjustment strategy is set as: In the formula Indicates the The step size for the iteration, Indicates the preset initial maximum step size, , represents the hyperbolic secant function, represents the decay rate factor, represents the index of the iteration number, and , Indicates the maximum number of iterations.

5. The method for evaluating shear strength of rock mass structural surfaces based on a neural network according to claim 1, wherein: The logic of using the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural surface is: The morphological characteristic parameters, mechanical data and correction factors are input into the multi-scale neural network model to extract the macroscopic morphological characteristic vector and the microscopic relief characteristic vector respectively. The morphological feature vector and the relief feature vector are processed through the multi-scale feature fusion layer to extract the fused feature vector, which is expressed as: In the formula 、 、 Represent the fusion feature vector, the morphology feature vector, and the undulation feature vector respectively, represents the weight matrix of the multi-scale feature fusion layer, represents the activation function, represents the feature concatenation operation, Represents the bias of the multi-scale feature fusion layer; Then, the predicted value of shear strength is calculated by linear mapping in the output layer, and the expression is: In the formula represents the predicted value of shear strength, 、 denote the weight matrix and bias of the output layer respectively.

6. The method for evaluating shear strength of rock mass structural surfaces based on a neural network according to claim 5, characterized in that: The logic for generating corresponding intensity level labels based on statistical distribution characteristics is as follows: Collect three-dimensional morphological data and mechanical data of different rock mass structural surfaces, and calculate the corresponding predicted value of shear strength; The predicted values ​​of shear strength of different rock mass structural planes are used as historical data, and their mean and standard deviation are calculated; A dynamic grading function is constructed to output the strength grade label corresponding to the predicted value of shear strength. The dynamic grading function expression is: In the formula Indicates the grade strength label, Indicates the preset reference value of shear strength, represents the standard deviation of historical data, represents the volatility factor, , Represents the clipping function, constraint , Indicates rounding down within the brackets; Grade Strength Label , and its confidence interval is marked as ,in , the predicted value of shear strength is proportional to the value of the grade strength label.

Images