Rock mass structural surface shear strength evaluation method based on neural network

Abstract

The invention provides a rock mass structural surface shear strength evaluation method based on a neural network, relates to the technical field of rock mass shear strength evaluation, and improves the precision of rock mass structural surface shear strength evaluation by integrating three-dimensional morphology analysis and multi-scale neural network modeling. The directivity weight matrix quantifies the influence of the structural plane direction heterogeneity on the shear strength, and the problem of insufficient characterization of the multi-scale morphology is avoided in combination with the surface fluctuation parameter and the equivalent compressive strength formula. Secondly, a dynamic correction factor and a self-adaptive weight optimization algorithm are introduced, the difference between the size effect and the material sensitivity is effectively compensated, and the model is prevented from falling into local optimum. A multi-scale feature fusion mechanism synchronously extracts a macroscopic morphology rule and microscopic fluctuation details, and an intensity grade label is generated through a dynamic grading function, so that the engineering decision complexity is simplified. Finally, the three-dimensional visual space mapping visually shows the association between the structure surface morphology and the strength distribution, and the interpretability of the result is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of rock mass shear strength evaluation, and particularly to a method for evaluating the shear strength of rock mass structural planes based on a neural network. Background Art

[0002] The evaluation of the shear strength of rock mass structural planes is a core link in the stability analysis of geotechnical engineering. Traditional methods mostly rely on empirical formulas or single mechanical parameters (such as normal stress, internal friction angle) for estimation, which have significant limitations. In the prior art, the harmonic mean method is often used for the equivalent strength calculation of non-uniform rock masses, but the non-linear influence of the strength difference between the hanging wall and the footwall on the shear behavior is ignored, resulting in a systematic underestimation of the strength of asymmetric rock masses. At the same time, the three-dimensional topographic features of rock mass structural planes (such as directional distribution, surface undulation) are not fully quantified and are only characterized by simplified parameters such as the roughness coefficient, making it difficult to reflect the coupling effect of multi-scale topography on shear strength.

[0003] In the prior art, the one with the publication number CN105466790A discloses a method for evaluating the shear strength of rock mass structural planes with anisotropic characteristics. Its steps include: (1) obtaining the three-dimensional topographic data of the rock mass structural plane; (2) calculating the anisotropic characteristic parameter SRv and the undulation amplitude parameter A of the structural plane, and using the two parameters SRv and A to represent the roughness in each direction of the three-dimensional rock mass structural plane; (3) testing the wall strength JCS and the basic friction angle of the same type of rock mass structural plane; (4) substituting the structural plane 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 topographic parameters, unable to comprehensively characterize multi-scale topographic features, and without introducing a dynamic correction and adaptive optimization mechanism, 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 the background art section is only used to enhance the understanding of the background of the present disclosure, so it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

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

[0006] To achieve the above purpose, the present invention provides the following technical solution: A method for evaluating the shear strength of rock mass structural planes based on a neural network, the specific steps include: S1: Collect the three-dimensional topographic data and mechanical data of the rock mass structural plane, and respectively extract the topographic characteristic parameters and generate correction factors; S2: Construct a multi-scale neural network model, input the 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 plane, and generate corresponding strength grade labels in combination with the statistical distribution characteristics; S4: Correlate the three-dimensional morphological data, the predicted value of the shear strength and the strength grade labels to generate a visual spatial mapping.

[0007] Preferably, the morphological feature parameters include the directional distribution feature and the surface undulation parameter; Among them, the calculation method of the directional distribution feature is: ; In the formula represents the direction angle in the th direction, represents the directional weight matrix of the direction angle in the th direction, represents the dominant direction angle in the th direction, represents the weight coefficient in the th direction, , and , the subscript represents the direction index, and the superscript also represents the direction index, represents the total number of directions, represents the attenuation coefficient, and .

[0008] Preferably, the surface undulation parameters include the apparent dip angle statistic and the height distribution statistic; When collecting the three-dimensional morphological data and mechanical data of the rock mass structural plane, several groups of sampling points are set, and the apparent dip angle statistic is the mean and root mean square of the apparent dip angles of the sampling points; The height distribution statistic is the mean and root mean square of the relative heights of the sampling points.

[0009] Preferably, the mechanical data includes the normal stress, the basic internal friction angle, the uniaxial compressive strength of the upper and lower wall rocks of the rock mass structural plane, and the equivalent compressive strength; Among them, the equivalent compressive strength is calculated according to the uniaxial compressive strengths of the upper and lower wall rocks of the rock mass structural plane, and the calculation formula is: ; In the formula represents the equivalent compressive strength, , respectively represent the uniaxial compressive strengths of the upper and lower wall rocks of the rock mass structural plane, represents the reference coupling index, and , represents the difference correction factor, and its calculation method is: ; In the formula represents the material sensitivity coefficient, and .

[0010] Preferably, the calculation method of the correction factor is: ; In the formula represents the correction factor, represents the length of the rock mass structural plane, represents the preset reference length, , represents the adjustment factor of the formula, , , represents the non-linear compensation term, represents the sign function, and when the value in the parentheses is greater than 0, the function value is taken as 1, and when the value in the parentheses is less than or equal to 0, it is taken as -1; The calculation method of the non-linear compensation term is: ; In the formula represents the dynamic adjustment factor, and .

[0011] Preferably, the adaptive weight optimization algorithm adopts the dynamic adaptive cuckoo algorithm, and its step size adjustment strategy is set as: ; In the formula represents the step size at the th iteration, represents the preset initial maximum step size, , represents the hyperbolic secant function, represents the attenuation rate factor, represents the index of the iteration number, represents the maximum number of iterations, and .

[0012] 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 plane is: Input the morphological feature parameters, mechanical data and correction factor into the multi-scale neural network model, and extract the macroscopic morphological feature vector and the microscopic undulation feature vector respectively; Process the morphological feature vector and the undulation feature vector through the multi-scale feature fusion layer to extract the fused feature vector, and the expression is: ; wherein , , respectively represent the fused feature vector, the morphological feature vector, and the undulation feature vector, 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 the shear strength is calculated through a linear mapping in the output layer, and the expression is: ; wherein represents the predicted value of the shear strength, , respectively represent the weight matrix and the bias of the output layer.

[0013] Preferably, the logic for generating the corresponding strength level label by combining the statistical distribution characteristics is: Collect the three-dimensional morphological data and mechanical data of different rock mass structural planes, and calculate the predicted values of their corresponding shear strengths; Take the predicted values of the shear strengths of different rock mass structural planes as historical data, and statistically calculate their mean and standard deviation; Construct a dynamic grading function, and output the strength level label corresponding to the predicted value of the shear strength. The expression of the dynamic grading function is: ; wherein represents the grade strength label, represents the preset reference value of the shear strength, represents the standard deviation of the historical data, represents the fluctuation factor, , represents the clipping function, which constrains , represents rounding down the value inside the parentheses; The grade strength label , and its confidence interval is marked as , wherein , and the predicted value of the shear strength is proportional to the value of the grade strength label.

[0014] Compared with the prior art, the beneficial effects of the present invention are: The present invention significantly improves the accuracy and adaptability of the shear strength assessment of rock mass structural planes through the integration of three-dimensional topography analysis and multi-scale neural network modeling. First, the directional weight matrix quantifies the influence of the directional heterogeneity of the structural plane on the shear strength. Combining the surface undulation parameters with the equivalent compressive strength formula, it solves the problem of insufficient characterization of non-uniform rock masses and multi-scale topography by traditional methods. Second, the introduction of a dynamic correction factor and an adaptive weight optimization algorithm effectively compensates for the size effect and material sensitivity differences, preventing the model from falling into local optima. The multi-scale feature fusion mechanism synchronously extracts the macroscopic topography rules and microscopic undulation details, and generates intensity level labels through a dynamic grading function, simplifying the complexity of engineering decisions. Finally, the three-dimensional visualization space mapping intuitively presents the correlation between the structural plane topography and the strength distribution, enhancing the interpretability of the results. Brief Description of the Drawings

[0015] Figure 1 It is a schematic diagram of the overall method flow of the present invention. Detailed Embodiment

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with specific embodiments.

[0017] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those of ordinary skill in the field to which the present invention belongs. The "first", "second", and similar terms used in the present invention do not indicate any order, quantity, or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this term cover the elements or objects listed after this term and their equivalents, without excluding other elements or objects. The terms such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0018] Embodiment: Please refer to Figure 1 , the present invention provides a technical solution: A method for assessing the shear strength of rock mass structural planes based on a neural network, the specific steps including: S1: Collect the three-dimensional topography data and mechanical data of the rock mass structural plane, and respectively extract the topography feature parameters and generate correction factors.

[0019] The topography feature parameters include the directional distribution feature and the surface undulation parameter; Among them, the calculation method of the directional distribution feature is: ; In the formula, represents the direction angle in the th direction, represents the directional weight matrix of the direction angle in the th direction, represents the dominant direction angle in the th direction. The dominant direction angle refers to the dominant geometric direction distribution angle in the rock mass structural plane, which is usually consistent with the direction where shear sliding is likely to occur in the structural plane and is used to reflect the directional characteristics of the rock mass surface topography. When specifically setting it, the normal vector can be calculated for each local area first and converted into a direction angle, then the distribution frequency of all local direction angles is statistically analyzed, the angle interval with high frequency occurrence is identified, and finally algorithms such as K-means and DBSCAN are used to cluster the direction angle data, and the similar directions are classified into the same dominant direction. represents the weight coefficient in the th direction, and , and its specific setting method can be selected according to the actual situation: for example, by measuring the occurrence frequency of different direction angles in the rock mass structural plane and directly using the normalized frequency value as the weight coefficient; or performing principal component analysis on the structural plane direction data, extracting the main direction (the direction with the largest variance contribution), and allocating weights according to the variance contribution rate of the principal component; or directly setting scores based on expert experience, etc. The subscript represents the direction index, and the superscript also represents the direction index, represents the total number of directions, represents the attenuation coefficient, and .

[0020] Here, the calculation principle of the directional distribution characteristics is to calculate the contribution weight of different direction angles to the current analysis direction through the weight coefficient and the attenuation coefficient. The exponential term in the formula is used to represent the influence of the direction difference on the weight and can identify the dominant sliding direction of the rock mass structural plane, that is, the direction with lower shear strength.

[0021] When collecting the three-dimensional topography data and mechanical data of the rock mass structural plane, several groups of sampling points are set. When setting the sampling points, the entire area of the rock mass structural plane should be covered, including the areas with significant undulations (such as protrusions and depressions), the directional dominant areas (such as the dominant joint directions), and the transition areas (the areas with gradual topography changes), avoiding local dense or sparse sampling to ensure the spatial distribution uniformity, and its specific setting is determined according to the actual rock mass structural plane and will not be elaborated here. The surface undulation parameters include the apparent dip angle statistic and the height distribution statistic; The apparent dip angle statistic is the mean and root mean square of the apparent dip angles of the sampling points; The height distribution statistic is the mean and root mean square of the relative height of the sampling points.

[0022] The mechanical data includes normal stress, basic internal friction angle, uniaxial compressive strength of the rock walls on the upper and lower plates of the rock mass structural plane, and equivalent compressive strength. Here, the normal stress and basic internal friction angle can be obtained by conducting triaxial compression tests on the samples at the sampling points in the laboratory: for the normal stress, by applying normal loads under different confining pressures and inversely deducing based on the failure criterion (such as Mohr-Coulomb); for the basic internal friction angle, applying confining pressure to the sample, gradually increasing the axial stress until failure, and then plotting the Mohr circle, and it can be determined through the slope of the envelope line. The uniaxial compressive strength of the rock walls on the upper and lower plates of the rock mass structural plane can be obtained by uniaxial compression tests: after processing the samples collected from the upper and lower plate rock walls into standard specimens, applying axial load at a constant rate (such as 0.5 - 1 MPa / s), and recording the maximum stress at failure, which is the uniaxial compressive strength.

[0023] Among them, the equivalent compressive strength is calculated based on the uniaxial compressive strengths of the rock walls on the upper and lower plates of the rock mass structural plane, and the calculation formula is: ; In the formula represents the equivalent compressive strength, , respectively represent the uniaxial compressive strengths of the rock walls on the upper and lower plates of the rock mass structural plane, represents the reference coupling index, and , represents the difference correction factor, and its calculation method is: ; In the formula represents the material sensitivity coefficient, and .

[0024] It can be seen from the calculation formula of the equivalent compressive strength that its basic principle is based on the harmonic mean, and a difference correction factor is introduced to reflect the influence of the strength difference between the upper and lower plates on the equivalent strength. The basic term in the parentheses is used to reflect the equilibrium effect of the strengths of the upper and lower plates of the rock mass structure, and the exponential term is used to amplify the difference effect. When at this time , it increases the equivalent strength, otherwise it decreases, thus avoiding the underestimation of asymmetric strength by the traditional harmonic mean.

[0025] The calculation method of the correction factor is: ; In the formula represents the correction factor, represents the length of the rock mass structural plane, represents the preset reference length, , represents the adjustment factor of the formula, , and the values of both can be fitted through historical data or determined according to expert experience. represents the non - linear compensation term, represents the sign function, and when the value inside the parentheses is greater than 0, the function value is taken as 1, and when the value inside the parentheses is less than or equal to 0, it is taken as - 1; The calculation method of the non - linear compensation term is: ; In the formula represents the dynamic adjustment factor, and .

[0026] It can be seen from the calculation formula of the correction factor that , are respectively used to control the responses of the linear and quadratic terms to the length ratio, so as to adapt to the size effect at different scales. And is used to achieve dynamic adjustment of the compensation reverse, that is, compensates when , compensates when , thus reducing the non - linear error caused by length mutation and improving the calculation accuracy of the model.

[0027] In this step, by collecting the three - dimensional topography data and mechanical data of the rock mass structural plane, key characteristic parameters (appearance characteristic parameters and correction factors) are extracted, which can provide inputs for subsequent model training. Specifically, the three - dimensional topography data captures the macroscopic and microscopic geometric features of the structural plane (such as directional distribution, surface undulation), making up for the deficiencies of traditional single mechanical parameters. The correction factor solves the problem of the traditional model's neglect of the scale effect by introducing dynamic parameters such as the length of the structural plane, improving the prediction adaptability. The equivalent uniaxial compressive strength formula (JCS_K) of the mechanical data fuses the strength differences of the hanging wall and footwall rocks, enhancing the characterization ability of non - homogeneous rock masses.

[0028] S2: Construct a multi - scale neural network model, input the topography characteristic parameters, mechanical data and correction factors, and train and optimize through the adaptive weight optimization algorithm; In step S2, the adaptive weight optimization algorithm adopts the dynamic adaptive cuckoo algorithm, and its step - size adjustment strategy is set as: ; In the formula represents the step - size at the th iteration, represents the preset initial maximum step - size, , denotes the hyperbolic secant function, denotes the attenuation rate factor, denotes the index of the number of iterations, denotes the maximum number of iterations, and .

[0029] Specifically, the logic of the dynamically adaptive cuckoo algorithm here is as follows: In the initialization stage, multiple groups of neural network weight combinations (each combination is called a "bird's nest") are randomly created to cover different parameter configuration possibilities, and parameters such as the step size range and elimination probability are initialized, endowing the algorithm with the basic ability of adaptive adjustment. For example, the morphological feature parameters, mechanical data, and correction factors are used as input parameters to randomly generate multiple groups of weight combinations, that is, multiple "bird's nests", for mapping the input parameters to the predicted values of the shear strength; In the fitness evaluation stage, the shear strength is predicted on the validation set using each group of weights, and the error between the predicted value and the true value is used as the evaluation criterion (the smaller the error, the better the weights), and all weight combinations are sorted according to the error, and the current optimal solution is retained. It can be understood that the error here can be calculated using a variety of different methods, and the mean squared error is one of the most commonly used; In the Lévy search stage, for non-optimal weight combinations, they are updated according to the random pattern of Lévy flight, that is, long jumps avoid local optima, short step sizes perform fine searches, and the step size is automatically adjusted according to the population diversity. Referring to the calculation formula in the above step size adjustment strategy, the attenuation rate factor is affected by the population diversity. In the initial stage of iteration, the population diversity is large, and the attenuation rate factor is set to be smaller to perform global exploration with large step sizes, allowing individuals to be widely distributed in the search space. As the number of iterations increases, the population diversity slowly decreases, the attenuation rate factor is set to be larger, and the step size attenuation accelerates. The algorithm focuses on local area fine searches while improving the convergence efficiency. The attenuation rate factor can be dynamically set according to the number of iterations. For example, at , it is fixed at 0.1 to force global search, and at , it is set between 0.1 and 0.5 to balance the relationship between search and exploitation. In the last stage, it is set to 0.5 to accelerate convergence. It can also be set proportionally according to the relationship between and , which will not be elaborated here one by one; Then, monitor the dispersion degree of the population, calculate the distribution difference of all weight combinations (a large difference indicates high diversity). If the population is too concentrated, eliminate the poorly performing weight combinations according to the probability, and generate new solutions near the optimal solution to maintain the search vitality. Stop when the prediction error no longer decreases significantly (such as the change is less than the threshold for 5 consecutive iterations) or reaches the maximum number of iterations.

[0030] Since the dynamic adaptive cuckoo algorithm is a prior art, its specific calculation formula and method will not be elaborated here.

[0031] During the model training process, the measured values of the shear strength of the samples collected in the above steps can be measured and then input into the model for training. Specifically, the measured values of the shear strength can be obtained by applying shear force to the rock mass structural plane samples in the laboratory. Gradually apply the shear force until the structural plane slides, and record the maximum shear stress as its shear strength. Divide the data of the sampling points of the rock mass structural plane into an 80% training set and a 20% validation set. And use the mean square error (MSE) to measure the difference between the predicted value and the measured value of the shear strength, and then use the backpropagation algorithm to update the model. If the mean square error is less than the preset convergence threshold in several consecutive iterations of the model, it is considered that the model optimization is completed, and the specific number of times can be determined according to expert experience. It is also possible to set the maximum number of iterations, force convergence when the maximum number of iterations is reached, and take the historical optimal weight (that is, the weight corresponding to the minimum mean square error).

[0032] In this step, a multi-scale neural network model is constructed and trained using the adaptive weight optimization algorithm (dynamic adaptive cuckoo algorithm) to achieve feature fusion and model optimization. Specifically, multi-scale feature extraction can capture both macroscopic morphology (such as directional distribution) and microscopic undulations (such as dip angle statistics) simultaneously, avoiding the limitations of single-scale features; adaptive weight optimization can balance global search and local convergence by dynamically adjusting the step size to prevent the model from falling into local optima; using the correction factor as an input parameter enhances the adaptability of the model to actual engineering conditions.

[0033] S3: Use the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural plane, and generate the corresponding strength grade label in combination with the statistical distribution characteristics.

[0034] 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 plane is as follows: Input the morphology feature parameters, mechanical data, and correction factor into the multi-scale neural network model, and extract the macroscopic morphology feature vector and microscopic undulation feature vector respectively; Process the morphology feature vector and undulation feature vector through the multi-scale feature fusion layer to extract the fused feature vector, and the expression is: ; In the formula , , represent the fused feature vector, morphology feature vector, and undulation feature vector respectively, represents the weight matrix of the multi-scale feature fusion layer, denotes the activation function, denotes the feature concatenation operation, denotes the bias of the multi-scale feature fusion layer; Then, the predicted value of the shear strength is calculated through a linear mapping in the output layer, and the expression is: ; In the formula denotes the predicted value of the shear strength, , respectively denote the weight matrix and bias of the output layer.

[0035] Specifically, the macroscopic morphological features can include the overall undulation trend (such as the main dip direction, average height distribution), regional structural features (such as the dominant orientation of joint groups, distribution of large protrusions and concavities), and the spatial distribution pattern of the directional weight matrix (polar coordinate statistical characteristics of the directional weight matrix), which are used to characterize the large-scale morphological laws of rock mass structural planes and can be extracted through a 5X5 convolution kernel in a multi-scale neural network; the microscopic undulation feature vectors can include the fluctuation of the standard deviation of the apparent dip angle, roughness details (such as micron-scale surface texture, fracture density), and local outliers of the directional weight matrix (i.e., mutation regions), which are used to characterize the local detail features of the structural plane and can be extracted through a 1X1 convolution kernel in a multi-scale neural network. The concatenation of macroscopic and microscopic features is essentially the complementarity of multi-scale information. Macroscopic features provide the overall strength trend (such as the influence of the dominant dip angle), and microscopic features correct local anomalies (such as strength attenuation caused by excessive roughness), thereby improving the accuracy of the model and avoiding the limitations of traditional single-scale models.

[0036] In step S3, the logic of generating the corresponding strength level label by combining statistical distribution features is as follows: Collect the three-dimensional morphological data and mechanical data of different rock mass structural planes, and calculate the predicted values of their corresponding shear strengths; Take the predicted values of the shear strengths of different rock mass structural planes as historical data, and statistically calculate their mean and standard deviation; Construct a dynamic grading function to output the strength level label corresponding to the predicted value of the shear strength. The expression of the dynamic grading function is: ; In the formula denotes the grade strength label, denotes the preset reference value of the shear strength, and its specific value can be determined according to expert experience or engineering standards, denotes the standard deviation of the historical data, denotes the fluctuation factor, , denotes the clipping function, which constrains , Denotes the floor function of the value inside the parentheses; Grade strength label , and its confidence interval is marked as , where , the predicted value of the shear strength is proportional to the value of the grade strength label.

[0037] Here, the predicted value of the shear strength is standardized by the mean and standard deviation of historical data. Each grade corresponds to a standard deviation interval, making it more intuitive and able to simplify engineering decisions.

[0038] In this step, multi-scale feature fusion is performed by splicing macroscopic and microscopic features, which can calculate the predicted value of the shear strength more accurately. Then, the mean and standard deviation of historical data are used to construct a grading rule to achieve adaptive updating of the classification results and provide risk warnings for engineering decisions.

[0039] S4: Associate the three-dimensional topography data, the predicted value of the shear strength, and the strength grade label to generate a visual spatial mapping.

[0040] In this step, the corresponding relationship between the structural plane topography and the strength grade is visually displayed through three-dimensional visualization, assisting engineers to quickly locate high-risk areas, enhancing the interpretability of the results, and facilitating comparison and verification with on-site measured data.

[0041] In summary, the present invention significantly improves the accuracy and adaptability of the shear strength assessment of rock mass structural planes by integrating three-dimensional topography analysis and multi-scale neural network modeling. First, the directional weight matrix quantifies the influence of the structural plane direction heterogeneity on the shear strength. Combining the surface undulation parameters and the equivalent compressive strength formula solves the problem of insufficient characterization of non-uniform rock masses and multi-scale topographies by traditional methods. Second, the introduction of a dynamic correction factor and an adaptive weight optimization algorithm effectively compensates for the size effect and material sensitivity differences, preventing the model from falling into a local optimum. The multi-scale feature fusion mechanism synchronously extracts macroscopic topography laws and microscopic undulation details, and generates strength grade labels through a dynamic grading function, simplifying the complexity of engineering decisions. Finally, the three-dimensional visual spatial mapping intuitively presents the relationship between the structural plane topography and the strength distribution, enhancing the interpretability of the results.

[0042] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0043] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any 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 realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0044] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0045] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application.

Claims

1. A method for evaluating the shear strength of rock mass structural planes based on a neural network, characterized in that, The specific steps include: S1: Collect the three-dimensional topography data and mechanical data of the rock mass structural plane, and obtain the topography characteristic parameters and correction factors based on this; S2: Construct a multi-scale neural network model, input the topography characteristic 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 plane, and generate the corresponding strength grade label in combination with the statistical distribution characteristics; S4: Associate the three-dimensional topography data, the predicted value of the shear strength and the strength grade label to generate a visual spatial mapping.

2. The shear strength evaluation method of rock mass structural plane based on neural network according to claim 1, characterized in that: The topography characteristic parameters include the directional distribution characteristic and the surface undulation parameter; Among them, the calculation method of the directional distribution characteristic is: ； wherein represents the direction angle in the th direction, represents the directivity weight matrix of the direction angle in the th direction, represents the dominant direction angle in the th direction, represents the weight coefficient in the th direction, and , where the subscript represents the direction index, and the superscript also represents the direction index, represents the total number of directions, represents the attenuation coefficient, and .

3. The shear strength evaluation method of rock mass structural plane based on neural network according to claim 2, wherein: The surface undulation parameter includes the apparent dip angle statistic and the height distribution statistic; When collecting the three-dimensional topography data and mechanical data of the rock mass structural plane, several groups of sampling points are set, and the apparent dip angle statistic is the mean and root mean square of the apparent dip angles of the sampling points; The height distribution statistic is the mean and root mean square of the relative heights of the sampling points.

4. A method for evaluating the shear strength of rock mass structural planes based on a neural network according to claim 1, characterized in that: The mechanical data includes the normal stress, the basic internal friction angle, the uniaxial compressive strength of the rock walls on the upper and lower plates of the rock mass structural plane, and the equivalent compressive strength; Among them, the equivalent compressive strength is calculated according to the uniaxial compressive strengths of the rock walls on the upper and lower plates of the rock mass structural plane, and the calculation formula is: ； In the formula represents the equivalent compressive strength , respectively represent the uniaxial compressive strengths of the rock walls above and below the rock mass structural plane represents the reference coupling index, and , represents the differential correction factor, and its calculation method is as follows: ； In the formula represents the material sensitivity coefficient, and .

5. The shear strength evaluation method of rock mass structural plane based on neural network according to claim 4, characterized in that: The calculation method of the correction factor is: ； wherein represents a correction factor, represents the length of the rock mass structural plane, represents a preset reference length, 、 represent the adjustment factors of the formula, , , represents a non-linear compensation term, represents the sign function, the function value is taken as 1 when the value in the parentheses is greater than 0, and -1 when the value in the parentheses is less than or equal to 0; The calculation method of the non-linear compensation term is: ； where represents a dynamic adjustment factor, and .

6. The shear strength evaluation method of rock mass structural plane based on neural network according to claim 1, characterized in that: The adaptive weight optimization algorithm adopts a dynamic adaptive cuckoo algorithm, and its step size adjustment strategy is set as: ； In the formula represents the step size at the -th iteration, represents the preset initial maximum step size, , represents the hyperbolic secant function, represents the attenuation rate factor, represents the index of the iteration number, and , represents the maximum number of iterations.

7. A method for evaluating the shear strength of rock mass structural planes based on a neural network according to claim 1, characterized in that: The logic of using the multi-scale neural network model to output the predicted value of the shear strength of the rock mass structural plane is: Input the topography characteristic parameters, mechanical data and correction factors into the multi-scale neural network model, and extract the macroscopic topography feature vector and the microscopic undulation feature vector respectively; Process the topography feature vector and the undulation feature vector through the multi-scale feature fusion layer to extract the fusion feature vector, and the expression is: ； In the formula , , represent the fused feature vector, the morphological 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 calculate the predicted value of the shear strength through linear mapping in the output layer, and the expression is: ； In the formula represents the predicted value of the shear strength, , respectively represent the weight matrix and bias of the output layer.

8. A method for evaluating the shear strength of rock mass discontinuities based on a neural network according to claim 7, characterized in that: The logic of generating the corresponding strength grade label in combination with the statistical distribution characteristics is: Collect the three-dimensional topography data and mechanical data of different rock mass structural planes, and calculate the corresponding predicted values of the shear strength; Take the predicted values of the shear strength of different rock mass structural planes as historical data, and statistically calculate their mean and standard deviation; Construct a dynamic grading function to output the strength grade label corresponding to the predicted value of the shear strength, and the expression of the dynamic grading function is: ； where represents the grade strength label, represents the preset reference value of the shear strength, represents the standard deviation of the historical data, represents the fluctuation factor, , represents the amplitude limiting function, which constrains , represents rounding down the value inside the parentheses; Grade strength label , whose confidence interval is labeled as , where , the predicted value of the shear strength is proportional to the value of the grade strength label.

Images