A rapid evaluation method for deep toppling stability of rock slopes

By constructing a position-constrained Bayesian optimization limit learning machine model and a dual-mode mechanical evaluation, combined with the scenario-adaptive coefficient of variation method, the problems of low accuracy and poor generalization in the evaluation of deep toppling stability of rock slopes were solved, and a fast and accurate multi-scenario adaptive evaluation was achieved.

CN123070201APending Publication Date: 2026-11-10CHINA HYDROELECTRIC ENGINEERING CONSULTING GROUP CHENGDU RESEARCH HYDROELECTRIC INVESTIGATION DESIGN AND INSTITUTE +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611294584.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for evaluating the stability of deep rock slopes are characterized by high cost, long cycle time, low accuracy, and poor generalization, making it difficult to meet the needs of regional batch investigations and rapid assessment in emergency scenarios.

Method used

A position-constrained Bayesian optimization extreme learning machine model is used to predict the depth of the bending zone. Combined with the slope rock strata deflection angle and crack grade correction, a dual-mode mechanical evaluation of cantilever beam tensile failure and bending zone shear is established. The scenario-adaptive coefficient of variation method is used to dynamically allocate weights and construct a rapid evaluation method.

Benefits of technology

It improves the accuracy of bending zone depth estimation, reduces the evaluation error caused by a single failure criterion, enhances the method's relevance and rationality in special areas, and achieves convenient parameter acquisition and adaptability to multiple scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070201A_ABST
    Figure CN123070201A_ABST
Patent Text Reader

Abstract

This invention discloses a rapid evaluation method for the stability of deep toppling rock slopes, belonging to the technical field of rock slope stability evaluation. Addressing the pain points of existing deep toppling assessment methods that rely on drilling, are time-consuming, and costly, this method collects basic parameters such as slope height, rock stratum attitude, lithology, and deformation characteristics using non-contact means. It then uses a pre-trained attitude-constrained Bayesian optimization limit learning machine model to predict the depth of the bend zone, and introduces correction coefficients for slope rock stratum deflection angle and crack level for correction. Based on this, it calculates the cantilever beam tensile safety factor and the bend zone shear safety factor to obtain a comprehensive mechanical safety factor. Finally, it uses a scenario-adaptive coefficient of variation method to dynamically allocate weights for mechanical, deformation, and apparent indicators, weighting them to obtain a comprehensive stability coefficient and determine the stability level. This invention achieves rapid multi-indicator fusion evaluation of deep toppling deformation, with easily obtainable parameters and reliable evaluation results, making it suitable for regional slope disaster screening.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the technical field of geological hazard risk assessment for rock slopes, specifically relating to a rapid evaluation method for the deep toppling stability of rock slopes. Background Technology

[0002] Reverse-dip layered rock slopes are a widely distributed slope type in mountainous areas. Deep overturning is a unique progressive failure mode of this type of slope. Its deformation core is hidden inside the mountain, and the surface only shows sporadic cracks. The precursory characteristics before instability are extremely inconspicuous. Once instability occurs, the scale of damage can reach millions of cubic meters, which can easily trigger major geological disasters such as blocking rivers, destroying roads, and burying residential areas. Therefore, it is necessary to carry out extensive and efficient stability assessments regularly.

[0003] Currently, the evaluation methods for the deep toppling stability of rock slopes are mainly divided into two categories: one is the traditional fine evaluation method, which obtains the depth of the bending zone through drilling, conducts indoor mechanical tests by taking samples on site, and calculates the safety factor by combining numerical simulation or limit equilibrium method. Although this type of method has high evaluation accuracy, the evaluation cost of a single slope is high, the evaluation cycle is long, and drilling operations are difficult to carry out in remote mountainous areas with complex terrain. It is only suitable for the exploration scenarios of a single key project and cannot support the needs of regional batch investigation and rapid assessment of emergency scenarios.

[0004] Another type is the existing rapid evaluation method, which mainly estimates the depth of the bend zone based on empirical formulas and judges stability by combining apparent crack characteristics. Although this type of method does not require drilling, it has three obvious limitations: First, the depth of the bend zone is estimated empirically only through a few parameters such as slope height and rock dip angle, without correction by surface toppling deformation characteristics, resulting in insufficient estimation accuracy in deep toppling scenarios; Second, traditional mechanical calculations mostly use a single failure criterion, failing to effectively integrate the combined effects of the two instability mechanisms of cantilever beam tensile failure and bend zone shear, leading to deviations between the mechanical evaluation results and the actual failure mode; Third, the evaluation indicators use fixed weight allocation, failing to consider the regional disturbance effects in high-intensity earthquake zones and heavy rainfall zones, resulting in poor generalization in special areas and failing to meet the evaluation needs of multiple scenarios.

[0005] In summary, there is an urgent need to propose a rapid evaluation method for deep slope toppling stability that is easy to obtain, can take into account multiple failure mechanisms, and has adaptive and dynamic weight adjustment, so as to support the screening and early warning of regional geological disasters. Summary of the Invention

[0006] To address the problems existing in the background technology, this invention proposes a rapid evaluation method for the stability of deep toppling rock slopes. This method utilizes a bend zone depth prediction mechanism based on a state-constrained Bayesian optimization limit learning machine, a collaborative correction mechanism for deflection angle and crack level, a mechanical evaluation mechanism fusing cantilever beam tensile failure and bend zone shear modes, and a dynamic weight allocation mechanism based on the scene-adaptive coefficient of variation method. These mechanisms solve core technical problems such as insufficient accuracy in bend zone depth estimation, large deviations between mechanical evaluation results and actual failure modes due to a single failure criterion, and poor generalization of fixed weights across different regional scenarios. The method achieves joint optimization of the convenience of parameter acquisition, accuracy of bend zone depth estimation, rationality of mechanical evaluation, and adaptability to multiple scenarios in deep toppling stability evaluation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A rapid evaluation method for the deep toppling stability of rock slopes includes the following steps: S1. Non-contact data acquisition of slope height, slope angle, primary rock layer dip angle, single-layer rock layer thickness, lithology code, slope rock layer deflection angle, slope crack grade, annual average deformation rate and regional attribute labels of the slope to be evaluated. S2. Calculate the estimated depth of the bend zone using a pre-trained attitude-constrained Bayesian optimization extreme learning machine model. Then, correct the estimated depth of the bend zone by combining the slope rock layer deflection angle and the slope crack level to obtain the final bend zone depth. S3. Calculate the tensile safety factor and the shear safety factor of the cantilever beam based on the final bending zone depth, and calculate the comprehensive mechanical safety factor. S4. After normalizing the comprehensive mechanical safety factor, deformation rate index, and apparent characteristic index of the slope to be evaluated, the weight of the index is dynamically allocated using the scenario adaptive coefficient of variation method, and the comprehensive stability coefficient is obtained by weighting, and the corresponding stability level is output.

[0008] Specifically, the process of constructing the pre-trained attitude-constrained Bayesian optimization extreme learning machine model in step S2 includes: (1) Construct a single hidden layer feedforward neural network structure with 5 neurons in the input layer, corresponding to the input feature vector. ,in Because of the slope height, The slope angle is... The dip angle of the primary rock strata. The thickness of a single rock layer, Lithology encoding; number of hidden layer neurons is The output layer has one neuron, corresponding to the estimated depth of the bend. ; (2) Define the forward propagation rule, and the forward propagation formula is: ; in, Index of hidden layer neurons; For the first The weights from each hidden layer neuron to the output layer; The sigmoid activation function is expressed as follows: In the formula The slope parameter of the activation function; For the first The input weights of each hidden layer neuron. For the first Bias of hidden layer neurons; (3) The deep dumping test cases verified by drilling are divided into training set, validation set and test set according to a preset ratio. The input features of the training set are the input feature vector. The label is the measured depth of the bend in the case study; during training, a loss function with an embedded attitude consistency regularization term is used, and its expression is: ; in, For the total loss, For the regularization term of occurrence, the weighting coefficients are... To estimate the depth of the bend zone in the model output after adding the attitude consistency regularization term, This is the initial estimate of the bend depth output by the model without the addition of the attitude consistency regularization term. This represents the deflection angle of the slope rock strata corresponding to the sample. The root mean square error of the validation set is calculated using the following formula: ; In the formula For the validation set sample index; The total number of samples in the validation set; For the first The model outputs a predicted depth of the bend in the test sample. For the first The measured bending depth of a verification sample; (4) Use Bayesian optimization algorithm to search for optimal hyperparameters , The optimization objective is the total loss. Minimum.

[0009] Specifically, the calculation formula for correcting the estimated depth of the bend in step S2 is as follows: ; in, This refers to the final bending depth; The estimated value of the bend depth output by the pre-trained model; This is the correction coefficient for the deflection angle of the slope rock strata. The slope rock strata deflection angle collected in step S1; This is the correction factor for the slope crack level.

[0010] Specifically, the cantilever beam tensile safety factor mentioned in step S3 The calculation formula is: ; in For the tensile strength of rock, To account for the maximum tensile stress after correction for interlaminar friction effect, the calculation formula is as follows: ; in, The rock is of high density. The dip angle of the primary rock strata collected in step S1. The thickness of the single rock layer collected in step S1. The coefficient of friction between rock layers. This is the length of the cantilever beam along the direction of the original rock strata.

[0011] Specifically, the shear safety factor of the bending strip mentioned in step S3 The calculation formula is: ; in, This is the regional disturbance reduction factor. For the effective cohesion of the rock mass in the bend zone, The rock is of high density. The final bending depth obtained in step S2 The slope inclination angle collected in step S1, This represents the friction angle within the rock mass of the bend zone.

[0012] Specifically, the comprehensive mechanical safety factor mentioned in step S3 The calculation formula is: ; in, The tensile safety factor of the cantilever beam calculated in step S3 is... The shear safety factor of the bent strip calculated in step S3. The final bending depth obtained in step S2 The slope height is the slope to be evaluated collected in step S1.

[0013] Specifically, the normalization rules in step S4 include: (1) Deformation rate index Normalization: ; in The annual average deformation rate collected in step S1; (2) Apparent characteristic indicators Normalization: Assign values ​​to the slope crack levels collected in step S1.

[0014] Specifically, the step of dynamically allocating index weights using the scenario-adaptive coefficient of variation method in step S4 includes: (1) Based on the regional attribute labels collected in step S1, match the statistical sample pool of publicly recorded deep dumping cases in the corresponding region, and obtain the sample standard deviation of the three normalized indicators. Sample mean ,in The indexes correspond to the comprehensive mechanical safety factor, deformation rate index, and apparent characteristic index, respectively. (2) Calculate the coefficient of variation for each indicator. : ; (3) Calculate the initial weights of each indicator. : ; When any indicator data is missing, the weight of that indicator is proportionally allocated to the remaining valid indicators.

[0015] Specifically, the formula for calculating the comprehensive stability coefficient in step S4 is as follows: ; in, The comprehensive stability coefficient of the slope to be evaluated is... For the first The index values ​​after normalization; The stability level of the slope to be evaluated is based on the comprehensive stability coefficient. The value range is divided as follows: when When, it is determined to be stable; when When, it is determined to be basically stable; when When, it is judged as unstable; when When the condition is met, it is determined to be unstable.

[0016] In summary, the beneficial technical effects of the present invention are as follows: 1. Improved accuracy of bending zone depth estimation: This invention constructs a position-constrained Bayesian optimization extreme learning machine model and embeds a position consistency regularization term into the loss function. It introduces the slope rock layer deflection angle into the model training process, so that the bending zone depth prediction model follows the geomechanical laws of toppling deformation while fitting the measured data. At the same time, it combines a dual correction mechanism of slope rock layer deflection angle and crack level, which overcomes the shortcomings of the existing empirical formula method, which estimates the depth based on only a few parameters such as slope height and rock layer dip angle, resulting in insufficient accuracy in deep toppling scenarios. This improves the accuracy of bending zone depth prediction. 2. Reduced evaluation error caused by single failure criterion: This invention establishes a dual-mode mechanical calculation system with a cantilever beam tensile failure safety factor and a bending zone shear safety factor, and dynamically adjusts the weight distribution of the two failure modes according to the ratio of bending zone depth to slope height. This effectively integrates the two instability mechanisms of cantilever beam tensile failure and bending zone shear, solves the problem of large deviation between the evaluation results of a single failure criterion and the actual failure mode, and makes the mechanical evaluation more consistent with the progressive failure characteristics of deep toppling. 3. Improved the relevance and rationality of the evaluation method in special areas: This invention introduces peak ground acceleration reduction in high-intensity earthquake zones and rainfall reduction in heavy rainfall zones by setting regional disturbance reduction coefficients, so that mechanical calculations can reflect the differentiated effects of environmental loads on the shear strength of bending zones in different regions; 4. Avoids the generalization defects caused by fixed weight allocation: This invention uses the scenario-adaptive coefficient of variation method to match the statistical sample pool of actual cases in the corresponding region based on the regional attribute labels. It dynamically allocates the weights of the three types of indicators, namely mechanics, deformation, and appearance, based on the coefficient of variation of each indicator in the same region. This solves the problem of poor generalization in different regional scenarios caused by the use of fixed empirical weights in existing methods. It realizes the adaptive adjustment of weights and makes the comprehensive evaluation results more in line with the objective statistical laws of each region. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] To make the technical means, creative features, objectives and effects of this invention clearer and easier to understand, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0019] Example This embodiment uses a reverse-dip layered rock slope in the Hengduan Mountains of southwestern my country as the object of evaluation. Figure 1 As shown, the present invention provides a rapid evaluation method for the deep toppling stability of rock slopes, which specifically includes the following steps: First, determine whether the slope to be evaluated meets the necessary conditions for deep-layer toppling development. If any condition is not met, the evaluation is terminated. The specific judgment rules and procedures are as follows: S01. Slope Angle Determination: The slope grid gradient of the slope is extracted from 30m resolution public SRTM DEM data. The average slope of the area with a slope coverage of more than 80% is used as the slope angle. ,determination Is it not lower than ; S02. Determination of Rock Strata Attitude: Obtain the attitude (strike, dip, and dip angle) of the primary rock strata in the area to which the slope belongs using a 1:200,000 publicly available regional geological map. Determine whether the angle between the rock strata dip and the slope dip is greater than 150°, i.e., whether it is a reverse-dip layered structure, and the dip angle of the primary rock strata. ; S03. Determination of Single-Layer Rock Thickness: Identify the spacing between bedding planes using Sentinel-2 satellite imagery or UAV aerial photography, and calculate the average spacing of more than 10 continuous bedding planes as the thickness of the single-layer rock layer. ,determination Is it no greater than 0.5m?

[0020] The slope will be evaluated only if all of the above conditions are met: S1. Non-contact basic parameter acquisition; The elevation difference between the slope toe and slope crest is extracted from SRTM DEM data as the slope height. The slope angle is calculated according to steps S01-S03 above. dip angle of primary rock strata Thickness of a single rock layer Lithology codes were matched using stratigraphic and lithological descriptions from a 1:200,000 publicly available regional geological map. The coding rule was: if it is limestone, then... Sandstone Slate / shale Phyllite The measured dip angle of the top layer of fallen rock on the slope was extracted using UAV LiDAR point cloud or close-range photogrammetry point cloud. Calculate the deflection angle of the rock strata on the slope. If point cloud data is unavailable, the difference in attitude between surface and primary rock layers can be estimated using high-resolution satellite imagery. Slope cracks can be visually identified using 0.1m resolution UAV aerial photography or 0.5m resolution Gaofen-7 satellite imagery, and graded according to the following standards: Level 1 (no visible cracks), Level 2 (discontinuous parallel cracks, single length and width), Level 3 (continuous parallel cracks, single length and width), Level 4 (step-like displacement or localized spalling). Using publicly available Sentinel-1 C-band SAR data, the annual average deformation rate of the slope surface can be extracted using the open-source SBAS-InSAR tool. The regional seismic intensity and multi-year average rainfall were obtained through publicly available data, and the regional attribute labels were obtained according to the following standards: Category A (high-intensity seismic zone, intensity VIII and above), Category B (heavy rainfall zone, annual rainfall greater than 1200 mm), and Category C (ordinary tectonic zone).

[0021] S2, Depth estimation of the bend zone; The estimated depth of the bend zone is calculated by using a pre-trained attitude-constrained Bayesian optimization extreme learning machine model. The estimated depth of the bend zone is then corrected by combining the slope rock layer deflection angle and the slope crack level to obtain the final bend zone depth. The construction process of the pre-trained attitude-constrained Bayesian optimization extreme learning machine model includes: (1) Construct a single hidden layer feedforward neural network structure with 5 neurons in the input layer, corresponding to the input feature vector. ,in Because of the slope height, The slope angle is... The dip angle of the primary rock strata. The thickness of a single rock layer, Lithology encoding; number of hidden layer neurons is The output layer has one neuron, corresponding to the estimated depth of the bend. ; (2) Define the forward propagation rule, and the forward propagation formula is: ; in, Index of hidden layer neurons; For the first The weights from each hidden layer neuron to the output layer; The sigmoid activation function is expressed as follows: In the formula The slope parameter of the activation function; For the first The input weights of each hidden layer neuron. For the first Bias of hidden layer neurons; (3) Collect publicly available, drill-verified deep-layer dumping test cases and divide them into training set, validation set and test set in a ratio of 7:2:1. The input features of the training set are the input feature vector. The label is the measured depth of the bend in the case study; during training, a loss function with an embedded attitude consistency regularization term is used, and its expression is: ; in, For the total loss, The weighting coefficient for the occurrence consistency regularization term is 0.2, which is the optimal fixed value selected after 5-fold cross-validation. To estimate the depth of the bend zone in the model output after adding the attitude consistency regularization term, This is the initial estimate of the bend depth output by the model without the addition of the attitude consistency regularization term. This represents the deflection angle of the slope rock strata corresponding to the sample. The root mean square error of the validation set is calculated using the following formula: ; In the formula For the validation set sample index; The total number of samples in the validation set; For the first The model outputs a predicted depth of the bend in the sample; For the first The measured bending depth of a verification sample; (4) Use Bayesian optimization algorithm to search for optimal hyperparameters , ,in The search range is integers in the range [10, 50]. The search range is floating-point numbers in the range [0.1, 10], and the optimization objective is the total loss. The minimum value is obtained after 50 iterations, yielding the optimal hyperparameters. Complete the model pre-training.

[0022] Then, the collected input feature vectors of the slope to be evaluated are input into the pre-trained attitude-constrained BO-ELM model, and the output is the estimated value of the bend depth. ; Estimated depth of the bend The correction is performed using the following formula: ; in, This refers to the final bending depth; This is the correction coefficient for the deflection angle of the slope rock strata. The slope rock strata deflection angle collected in step S1; The slope crack level correction coefficient is determined based on the slope crack level classification collected in step S1: 0.95 for level 1, 1.0 for level 2, 1.1 for level 3, and 1.15 for level 4.

[0023] Subsequently, geophysical boundary verification was performed: the surface toppling thickness of the slope to be evaluated was obtained by identifying the distribution depth of the deflected rock layers using UAV point cloud data. That is, the vertical thickness of the surface rock mass where the slope deflects, if corrected Then take Ultimately, the final depth of the bend zone, consistent with geological patterns, was obtained. .

[0024] S3. Calculation of comprehensive mechanical safety factor using dual-control limit equilibrium model; Based on the final bending zone depth Calculate the tensile safety factor and the bending shear safety factor of the cantilever beam respectively, and calculate the comprehensive mechanical safety factor; Safety factor for tensile failure of cantilever beam The calculation formula is: ; in The tensile strength of the rock is determined by the lithology code obtained in step S1. Tiered value selection: Take 2.0, Take 1.5, Take 0.5, Take 0.3; To account for the maximum tensile stress after correction for interlaminar friction effect, the calculation formula is as follows: ; in, The rock is of high density. The dip angle of the primary rock strata collected in step S1. The thickness of the single rock layer collected in step S1. The friction coefficient between rock layers is determined by lithological coding. Graded value selection, Take 0.6, Take 0.55, Take 0.45, Take 0.4, This is the length of the cantilever beam along the direction of the original rock strata.

[0025] Bending belt shear safety factor The calculation formula is: ; in, The regional disturbance reduction factor is determined based on the regional attribute labels collected in step S1: Category B uses a rainfall reduction value of 0.9, and Category A uses an earthquake reduction value. ,in For regional peak ground acceleration, take 1.0 for Class C. To determine the effective cohesion of the folded zone rock mass, based on the lithological codes collected in step S1. Tiered value selection: Take 150, Take 100, Take 50, Take 30; The rock is of high density. The final bending depth obtained in step S2 The slope inclination angle collected in step S1, The friction angle within the folded zone is determined by the lithological code. Take 38, Take 35, Take 30, Take 27.

[0026] Furthermore, the dominant failure mode is determined based on the ratio of the bend depth to the slope height. and The comprehensive mechanical safety factor is obtained by performing segmented weighting. The calculation formula is: ; S4. Stability determination of multi-indicator fusion; After normalizing the comprehensive mechanical safety factor, deformation rate index, and apparent characteristic index of the slope to be evaluated, the weights of the indexes are dynamically allocated using the scenario adaptive coefficient of variation method, and the comprehensive stability coefficient is obtained by weighting, and the corresponding stability level is output.

[0027] The rules for normalization specifically include: (1) Deformation rate index Normalization: ; in The annual average deformation rate collected in step S1; (2) Apparent characteristic indicators Normalization: Assign values ​​according to the slope crack level collected in step S1: 1.2 for level 1, 1.0 for level 2, 0.8 for level 3, and 0.6 for level 4.

[0028] Then, the coefficient of variation method is used to dynamically allocate indicator weights based on regional statistical characteristics. The larger the coefficient of variation, the higher the discrimination of the indicator, and the greater the corresponding weight. The specific steps are as follows: (1) Based on the regional attribute labels collected in step S1, match the statistical sample pool of publicly recorded deep dumping cases in the corresponding region, and obtain the sample standard deviation of the three normalized indicators. Sample mean ,in The indexes correspond to the comprehensive mechanical safety factor, deformation rate index, and apparent characteristic index, respectively. (2) Calculate the coefficient of variation for each indicator. : ; (3) Calculate the initial weights of each indicator. : ; If data for a certain indicator is missing, the initial weight of that indicator is allocated to the remaining valid indicators according to their initial weight ratios. For example, if the deformation rate indicator is missing... 2) When the weights are redistributed, the weights are: ; When other indicators are missing, the weights after redistribution are calculated using the same logic.

[0029] Final calculation of the overall stability coefficient The calculation formula is: ; in, For the first The index values ​​after normalization; The stability level of the slope to be evaluated is then determined based on the comprehensive stability coefficient. The value range is divided as follows: when At that time, it was determined to be stable, with no risk of instability, and could be developed and utilized normally; when At this point, it is determined to be basically stable, and deformation monitoring needs to be carried out once every six months; when When the situation is deemed unstable, measures such as anchoring and slope cutting are required. when If the situation is deemed unstable, it is necessary to immediately evacuate threatened personnel and carry out emergency response.

[0030] Therefore, this invention provides a rapid evaluation method for the stability of deep toppling rock slopes. It overcomes the limitations of deep toppling evaluation, which relies on drilling and is unsuitable for remote areas without exploration conditions, through a fully non-contact parameter acquisition architecture. A customized BO-ELM model with embedded geological prior regularization terms addresses the problems of hidden and difficult-to-measure depth of bends and low accuracy of empirical estimations. A dual-control mechanical model with interlayer friction correction matches the progressive failure mechanism of deep toppling. An adaptive dynamic weight fusion mechanism adapts to the evaluation needs of different regions and data qualities. This constructs a rapid evaluation closed loop encompassing non-contact acquisition, bend depth estimation, mechanical reliability calculation, and multi-index fusion and grading. It solves the pain points of existing technologies, such as high evaluation costs and long cycles, low accuracy in identifying hidden deep deformations, and poor generalization of fixed weights, making them unsuitable for various scenarios. This significantly reduces the evaluation cycle and cost while ensuring evaluation accuracy, providing a low-cost, highly reliable, and universally applicable technical solution for large-scale geological disaster investigation in mountainous areas, emergency risk assessment, and preliminary exploration for transportation and water conservancy projects.

[0031] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A rapid evaluation method for the deep toppling stability of rock slopes, characterized in that, Includes the following steps: S1. Non-contact data acquisition of slope height, slope angle, primary rock layer dip angle, single-layer rock layer thickness, lithology code, slope rock layer deflection angle, slope crack grade, annual average deformation rate and regional attribute labels of the slope to be evaluated. S2. Calculate the estimated depth of the bend zone using a pre-trained attitude-constrained Bayesian optimization extreme learning machine model. Then, correct the estimated depth of the bend zone by combining the slope rock layer deflection angle and the slope crack level to obtain the final bend zone depth. S3. Calculate the tensile safety factor and the shear safety factor of the cantilever beam based on the final bending zone depth, and calculate the comprehensive mechanical safety factor. S4. After normalizing the comprehensive mechanical safety factor, deformation rate index, and apparent characteristic index of the slope to be evaluated, the weight of the index is dynamically allocated using the scenario adaptive coefficient of variation method, and the comprehensive stability coefficient is obtained by weighting, and the corresponding stability level is output.

2. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The process of constructing the pre-trained attitude-constrained Bayesian optimization extreme learning machine model in step S2 includes: (1) Construct a single hidden layer feedforward neural network structure with 5 neurons in the input layer, corresponding to the input feature vector. ,in Because of the slope height, The slope angle is... The dip angle of the primary rock strata. The thickness of a single rock layer, Lithology encoding; number of hidden layer neurons is The output layer has one neuron, corresponding to the estimated depth of the bend. ; (2) Define the forward propagation rule, and the forward propagation formula is: ; in, Index of hidden layer neurons; For the first The weights of each hidden layer neuron to the output layer; The sigmoid activation function is expressed as follows: In the formula The slope parameter of the activation function; For the first The input weights of each hidden layer neuron. For the first Bias of hidden layer neurons; (3) The deep dumping test cases verified by drilling are divided into training set, validation set and test set according to a preset ratio. The input features of the training set are the input feature vector. The label is the measured depth of the bend in the case study; during training, a loss function with an embedded attitude consistency regularization term is used, and its expression is: ; in, For the total loss, For the regularization term of occurrence, the weighting coefficients are... To estimate the depth of the bend zone in the model output after adding the attitude consistency regularization term, This is the initial estimate of the bend depth output by the model without the addition of the attitude consistency regularization term. This represents the deflection angle of the slope rock strata corresponding to the sample. The root mean square error of the validation set is calculated using the following formula: ; In the formula For the validation set sample index; The total number of samples in the validation set; For the first The model outputs a predicted depth of the bend in the sample; For the first The measured bending depth of a verification sample; (4) Use Bayesian optimization algorithm to search for optimal hyperparameters , The optimization objective is the total loss. Minimum.

3. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The formula for correcting the estimated depth of the bend in step S2 is as follows: ; in, This refers to the final bending depth; The estimated value of the bend depth output by the pre-trained model; This is the correction coefficient for the deflection angle of the rock strata on the slope. The slope rock strata deflection angle collected in step S1; This is the correction factor for the slope crack level.

4. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The safety factor for the tensile failure of the cantilever beam mentioned in step S3 The calculation formula is: ; in For the tensile strength of rock, To account for the maximum tensile stress after correction for interlaminar friction effect, the calculation formula is as follows: ; in, The rock is of high density. The dip angle of the primary rock strata collected in step S1. The thickness of the single rock layer collected in step S1. The coefficient of friction between rock layers. This is the length of the cantilever beam along the direction of the original rock strata.

5. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The shear safety factor of the bending strip mentioned in step S3 The calculation formula is: ; in, This is the regional disturbance reduction factor. For the effective cohesion of the rock mass in the bend zone, The rock is of high density. The final bending depth obtained in step S2 The slope inclination angle collected in step S1, This represents the friction angle within the rock mass of the bend zone.

6. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The comprehensive mechanical safety factor mentioned in step S3 The specific calculation formula is as follows: ; in, The tensile safety factor of the cantilever beam calculated in step S3 is... The shear safety factor of the bent strip calculated in step S3. The final bending depth obtained in step S2 The slope height of the slope to be evaluated is collected in step S1.

7. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The normalization rules in step S4 specifically include: (1) Deformation rate index Normalization: ; in The annual average deformation rate collected in step S1; (2) Apparent characteristic indicators Normalization: Assign values ​​to the slope crack levels collected in step S1.

8. The rapid evaluation method for deep toppling stability of rock slopes according to claim 1, characterized in that, The specific steps of dynamically allocating index weights using the scene-adaptive coefficient of variation method in step S4 are as follows: (1) Based on the regional attribute labels collected in step S1, match the statistical sample pool of publicly recorded deep dumping cases in the corresponding region, and obtain the sample standard deviation of the three normalized indicators. Sample mean ,in The indexes correspond to the comprehensive mechanical safety factor, deformation rate index, and apparent characteristic index, respectively. (2) Calculate the coefficient of variation for each indicator. : ; (3) Calculate the initial weights of each indicator. : ; When any indicator data is missing, the weight of that indicator is proportionally allocated to the remaining valid indicators.

9. A rapid evaluation method for deep toppling stability of rock slopes according to claim 8, characterized in that, The formula for calculating the comprehensive stability coefficient in step S4 is as follows: ; in, The comprehensive stability coefficient of the slope to be evaluated is... For the first The index values ​​after normalization; The stability level of the slope to be evaluated is based on the comprehensive stability coefficient. The value range is divided as follows: when When, it is determined to be stable; when When, it is determined to be basically stable; when When, it is judged as understability; when When the condition is met, it is determined to be unstable.