An Artificial Intelligence-Based Method for Searching Critical Slip Surfaces of Slopes

By employing an AI-based global coarse search and local fine search method, combined with finite element stress field and clustering algorithms, potential local extreme value regions are identified. This resolves the contradiction between global search capability and computational efficiency in existing technologies, enabling efficient and accurate search for critical sliding surfaces of slopes.

CN123069729APending Publication Date: 2026-11-10MCC CHENGDU RES INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610906519.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-06-23
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for searching critical sliding surfaces of slopes present a trade-off between global search capability and computational efficiency. Heuristic search methods offer good global coverage but are inefficient, while deterministic search methods are highly efficient but prone to getting trapped in local extrema.

Method used

An artificial intelligence-based approach is adopted, which involves finite element stress field calculation and safety factor function construction, combined with global coarse search and local fine search, using clustering algorithm to identify potential local extreme value regions, and then using deterministic method to perform fine search to determine the global critical sliding surface.

Benefits of technology

It improves computational efficiency, ensures the finding of the global optimum, avoids getting trapped in local extrema, reduces computation time by more than 50%, improves the accuracy of safety factor assessment by 10%-20%, and provides more comprehensive information for slope reinforcement design.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069729A_ABST
    Figure CN123069729A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of slope detection technology and provides an artificial intelligence-based method for searching the critical slip surface of slopes. It addresses the problem that existing technologies cannot simultaneously achieve both global search capability and computational efficiency. This invention combines the advantages of heuristic and deterministic methods, first using artificial intelligence algorithms to identify potential local extreme value regions, then employing deterministic methods for precise searching within these local regions, ultimately finding the global critical slip surface efficiently and accurately.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of slope detection technology, specifically relating to an artificial intelligence-based method for searching the critical sliding surface of a slope. Background Technology

[0002] Slope stability analysis is one of the core issues in geotechnical engineering, and the most critical task is to determine the critical slip surface (i.e., the potential slip surface with the minimum safety factor) and its corresponding minimum safety factor. Currently, critical slip surface search methods are mainly divided into two categories:

[0003] 1. Heuristic Search Methods

[0004] These methods do not require gradient information of the objective function with respect to the independent variables. Instead, they perform a global search by simulating mechanisms such as evolution and group behavior in nature. Typical examples include genetic algorithms, particle swarm optimization, and simulated annealing. The advantage of these methods is their strong global search capability; theoretically, they can find the global optimum. However, their disadvantage is low computational efficiency. To ensure the globality of the search, it is usually necessary to calculate tens of thousands of sliding surfaces, which is very time-consuming and difficult to apply to large-scale engineering or real-time analysis scenarios.

[0005] 2. Deterministic Search Methods

[0006] These methods rely on the gradient (or numerical approximation gradient) of the objective function with respect to the independent variables, iteratively searching for local minima along the negative gradient direction of the safety factor function. Typical examples include steepest descent, Directed Programming (DFP), and Newton's method. The advantages of these methods are fast convergence and high search efficiency; however, their disadvantages include a tendency to get trapped in local optima, meaning the search results heavily depend on the choice of the initial sliding surface and cannot guarantee finding the critical sliding surface corresponding to the globally minimum safety factor.

[0007] In summary, existing methods present a contradiction between "global search capability" and "computational efficiency." Heuristic search methods have good globality but low efficiency, while deterministic search methods are highly efficient but prone to getting trapped in local optima. Summary of the Invention

[0008] To address the problem that existing technologies cannot simultaneously achieve both global search capability and computational efficiency, this invention provides an artificial intelligence-based method for searching critical sliding surfaces of slopes. It combines the advantages of both heuristic and deterministic methods, first using artificial intelligence algorithms to identify potential local extreme value regions, and then employing deterministic methods to accurately search within these local regions, ultimately finding the global critical sliding surface efficiently and accurately.

[0009] To solve the technical problem, the technical solution adopted by this invention is as follows:

[0010] An artificial intelligence-based method for searching the critical slip surface of a slope includes:

[0011] S1: Finite element stress field calculation: Establish the slope geometric model, assign elastoplastic constitutive model parameters to the soil and rock mass, apply displacement boundary conditions and gravity loads, use the finite element method to calculate the internal stress field of the slope, and save the stress components of each element integration point.

[0012] S2: Sliding surface parameterization and safety factor function construction: The circular arc sliding surface is uniquely determined by three parameters: the x-coordinate of the sliding point X1, the x-coordinate of the sliding point X2, and the sag Y. The circular arc sliding surface is discretized, and the normal stress σ of each micro-segment is calculated by combining finite element stress field interpolation. n Based on the tangential stress τ, and the cumulative anti-slip force and sliding force of the circular arc, a safety factor function Fs=f(X1,X2,Y) is constructed. By inputting three parameters, the corresponding safety factor is obtained. This safety factor function serves as the objective function for subsequent optimization search.

[0013] S3: Global Coarse Search - Artificial Intelligence Identifies Potential Local Extrema Regions: Within feasible limits, generate at least 500 sets of random sliding surface parameters based on the three parameters: the x-coordinate of the entry point (X1), the x-coordinate of the exit point (X2), and the sag (Y). Then, call the constructed safety factor function to calculate the safety factor F for each set of sliding surface parameters. {s,i} After normalizing the sliding surface parameter set, a clustering algorithm is used to identify multiple local extrema regions with low safety factors, and the parameter subspace of each local extrema region is determined, along with the initial point P used as the subsequent deterministic fine search. init ;

[0014] S4: Local Fine-Search - Deterministic Method for Finding Local Minimum: Starting with an initial point P for each local minimum region. init Starting from point P, a gradient-based deterministic optimization algorithm is used to iteratively search for the local minimum safety factor in the corresponding parameter subspace, and a dual convergence criterion and boundary projection mechanism are set up. The current point P_best_local and the corresponding safety factor F_min_local are obtained as the accurate minimum safety factor and optimal sliding surface parameters of the local region.

[0015] S5: Determine the global critical sliding surface: Compare the local minima and take the one with the smallest safety factor as the global critical sliding surface and its minimum safety factor;

[0016] S6: Output critical sliding surface results: Output minimum safety factor, sliding surface geometric parameters, discrete trajectory and visualization results.

[0017] In some embodiments, the finite element stress field calculation in step S1 includes:

[0018] S11. Geometric Modeling: Establish a two-dimensional or three-dimensional geometric model based on the geometric dimensions of the slope;

[0019] S12. Assigning material parameters: Assigning elastic-plastic constitutive model parameters to the rock and soil mass, including density ρ, elastic modulus E, Poisson's ratio ν, cohesion c, internal friction angle φ, and dilatation angle ψ;

[0020] S13. Perform finite element mesh generation on the established geometric model;

[0021] S14. Boundary Conditions and Load Application: Apply displacement boundary conditions: fix all degrees of freedom at the bottom, constrain horizontal displacement on both sides; apply gravity load.

[0022] S15. Finite element solution: The Newton-Raphson iteration method is used for nonlinear elastoplastic solution to obtain the stress components at each element integration point. , , , , , The stress field results are saved as a data file for subsequent safety factor calculations.

[0023] In some embodiments, the sliding surface parameterization and safety factor function construction in step S2 includes:

[0024] S21. Determine the parameterization of the sliding surface: For a circular sliding surface, three parameters are used to uniquely determine it: the x-coordinate of the sliding point. x-coordinate of the sliding point 、Ya Taka;

[0025] S22. Establish geometric constraints: Calculate the coordinates of the circle center based on the three parameters X1, X2, and Y using geometric relationships. , and radius R;

[0026] S23. Discretized Circular Arc: Divide the circular arc of the sliding surface into M equal segments, and record the coordinates of the midpoint of each segment. , ) and the arc length of the micro-segment Δ ;

[0027] S24. Stress field interpolation: For the coordinates of the midpoint of each micro-segment ( , The normal stress at that point is obtained by interpolation from the finite element stress field saved in step S15. and tangential stress τ.

[0028] S25. Calculate the safety factor; accumulate the anti-slip force and sliding force along the arc to obtain the safety factor. :

[0029] ;

[0030] Where c and φ are the cohesion and internal friction angle of the soil, respectively; Here are the grid numbers that the sliding surface traverses; the total number of grids traversed is... indivual; For the number Normal stress at the integration point within a mesh element; For the number Tangential stress at the integration point within the mesh element; For the sliding surface in numbered Grid length;

[0031] S26: Construct a safety factor function; encapsulate the above calculation process into a function Fs=f( , The function takes three parameters (Y, y) as input and outputs the corresponding safety coefficient; this function serves as the objective function for subsequent optimization searches.

[0032] In some embodiments, the local coarse search-artificial intelligence identification of potential local extremum regions in step S3 includes:

[0033] S31. Determine the parameter sampling range: Based on the slope geometry, set the feasible domain boundaries for X1, X2, and Y;

[0034] S32. Generate initial sample points: Within the feasible region, use Latin hypercube sampling or uniform random sampling to generate N sets of sliding surface parameters, where N≥500, denoted as ( , , ), i=1,2,...,N;

[0035] S33. Calculate the safety factor for sample points: For each set of sliding surface parameters, call the safety factor function Fs=f( from step S26). , ,Y), calculate the corresponding safety factor F {s,i} ;

[0036] S34. Constructing the dataset: Combine the three-dimensional parameters of the sample points and their corresponding safety factors into four-dimensional data points: ( , , F {s,i} );

[0037] S35. Data Normalization: Normalize each dimension of the four-dimensional data;

[0038] S36. Cluster Analysis: Cluster the normalized dataset using a density-based clustering algorithm;

[0039] S37. Extract clusters: Extract clusters with low safety factors from the clustering results. Each cluster represents a potential local extremum region.

[0040] S38. Determine the parameter subspace for each local extremum region; for each cluster, calculate the minimum and maximum values ​​of its parameters to form a three-dimensional rectangular subspace: [ , ]×[ , ]×[ , ].

[0041] S39. Select the initial point for the local search; within each cluster, select the sample point with the lowest safety factor, or select the cluster center point (mean of each dimension), as the initial point P for the subsequent deterministic fine search. init .

[0042] In some embodiments, the local fine-search-deterministic method for finding local minima in step S4 includes:

[0043] S401. Set local search parameters: Define the maximum number of iterations Tmax and convergence tolerance. and ;

[0044] S402. Initialize the current point: Let the current parameter vector P... curr =P init ;

[0045] S403. Calculate the safety factor of the current point: Call the function f(P) curr ) to obtain F curr ;

[0046] S404. Calculate the gradient of the objective function: The gradient ∇F(P) is approximated using the central difference method. curr ):

[0047] ;

[0048] The step size coefficient for the sliding point can be taken as the sliding point. 0.01-0.05 times the range of variation;

[0049] ;

[0050] The step size coefficient for the slip-out point can be taken as the slip-out point. 0.01-0.05 times the range of variation;

[0051] ;

[0052] The step size coefficient for the sag can be taken as the sag. 0.01-0.05 times the range of variation;

[0053] S405. Check convergence condition: If ||∇F|| < Or the parameter changes in two consecutive iterations < If the convergence occurs, proceed to step S412.

[0054] S406. Determine the search direction: Use the steepest descent method, with the search direction d = -∇F (negative gradient direction).

[0055] S407. One-dimensional line search for optimal step size: In direction d, solve for the step size α× such that F(P) curr Minimize +α·d);

[0056] S408. Update parameters: P new =P curr +α×·d;

[0057] S409. Boundary Constraint Handling: If P new If the parameter subspace boundary of the local region determined in step S38 is exceeded, then P will be... new Project back onto the boundary;

[0058] S410. Update iteration count: t = t + 1, if t > Tmax, then force stop;

[0059] S411. Repeated Iteration: Let P curr =P new Return to step S403;

[0060] S412. Record the local optimal solution: When convergence occurs, record the current point P_best_local and the corresponding safety factor F_min_local as the precise minimum safety factor and optimal sliding surface parameters for the local region.

[0061] In some embodiments, determining the global critical sliding surface in step S5 includes:

[0062] S51. Collect the optimal solutions for each local region: Summarize the (P_best_local, F_min_local) of each local extreme value region obtained in step S412 into a list:

[0063] S52. Compare global minimum values. Iterate through the list, find the entry with the smallest safety factor, and record its parameters P_global and F_min_global.

[0064] In some embodiments, the output of the critical sliding surface result in step S6 includes:

[0065] S61: Output minimum safety factor: Output F_min_global;

[0066] S62: Output critical sliding surface geometric parameters: Output the x-coordinate of the sliding point X1, the x-coordinate of the sliding point X2, and the sag Y, as well as the calculated center coordinates ( , ) and radius R.

[0067] S63: Output sliding surface trajectory: Generate a discrete point coordinate sequence of the arc based on the parameters ( , Export as a text file or DXF format.

[0068] S64: Visualization: Draw the critical sliding surface on the slope profile and label the safety factor value.

[0069] S65: Generate Search Report. Automatically generate a search process report, including: total number of calculations, convergence status of each local region, iteration history curves, and final results.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] First, an artificial intelligence algorithm (such as clustering) is used for a global coarse search to identify multiple potential local extrema regions; then, a deterministic method is used for a fine search in the local regions. This strategy organically integrates the global search capability of heuristic methods and the fast convergence capability of deterministic methods, overcoming their inherent shortcomings.

[0072] This invention directly uses the real stress field obtained from finite element calculations to integrate the anti-slip force and sliding force of the circular arc sliding surface. Compared with the inter-strip force assumption used in the traditional limit equilibrium method, the calculation results are more consistent with the actual stress distribution, thus improving the accuracy of safety factor assessment.

[0073] Instead of simply selecting the global minimum sample, a clustering algorithm is used to automatically identify multiple local minima from N random samples. This avoids missing true extreme regions due to insufficient sampling and provides multiple potential starting points for subsequent deterministic searches.

[0074] Deterministic optimization is performed on each identified local extremum region to ensure that each local valley can be accurately found. Finally, the global minimum is obtained by comparison, which solves the problem that a single deterministic method is prone to getting trapped in local extrema.

[0075] The circular arc sliding surface is uniquely determined by three parameters: the sliding point X1, the sliding point X2, and the sag Y. This transforms the geometric search problem into a low-dimensional (three-dimensional) parameter optimization problem, reducing the complexity of the search space.

[0076] In the local fine-grained search, a dual convergence criterion of gradient magnitude and parameter change, as well as a boundary projection mechanism, are set up to ensure the numerical stability and convergence reliability of the algorithm.

[0077] In summary, this invention employs a two-stage strategy of "first conducting a global coarse search to identify potential regions, and then conducting a local fine search," which ensures that no global optimal solution is overlooked (avoiding getting trapped in local extrema) while also avoiding the inefficiency of purely heuristic methods that require trial calculations of a massive number of sliding surfaces. Compared to traditional genetic algorithms, computation time can be reduced by more than 50%; compared to the steepest descent method alone, the success rate of finding the global minimum is increased from less than 60% to over 95% (based on numerical experiments using similar slope models).

[0078] This invention uses the stress field calculated by the finite element method to replace the interstrip force assumption of the traditional limit equilibrium method, which more realistically reflects the stress distribution inside the slope. In particular, for slopes with complex geological conditions (such as multi-layered soil, weak interlayers, and supporting structures), the accuracy of safety factor calculation can be improved by 10%-20%.

[0079] For slopes with multiple potential sliding surfaces (e.g., shallow and deep sliding coexisting), this invention automatically identifies multiple local extreme value regions through artificial intelligence clustering, and can simultaneously provide safety factors for multiple dangerous sliding surfaces, providing more comprehensive information for slope reinforcement design.

[0080] The global coarse search step in this invention uses random sampling and does not depend on the initial solution; the local fine search step uses gradient information, resulting in fast convergence. The overall algorithm is insensitive to initial parameter values ​​and has strong robustness.

[0081] This method can be directly integrated into existing slope finite element analysis software as a post-processing module, automatically completing the search for the critical slip surface without the need for repeated manual calculations, thus greatly improving the efficiency of engineering design. Attached Figure Description

[0082] Figure 1 This is a flowchart of an embodiment of the artificial intelligence-based slope critical sliding surface search method of the present invention;

[0083] Figure 2 A geometric schematic diagram of the parameterization of the circular arc sliding surface of the slope;

[0084] Figure 3 This is a geometric schematic diagram of the discretized circular arc of the slope's circular sliding surface. Detailed Implementation

[0085] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are some, but not all, of the embodiments of the present invention. The following embodiments are only for specific illustration of the implementation methods of the present invention and do not limit the scope of protection of the present invention.

[0086] In embodiments of the invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0087] Combined with appendix Figure 1 The artificial intelligence-based slope critical slip surface search method of the present invention includes:

[0088] S1: Finite element stress field calculation: Establish the slope geometric model, assign elastoplastic constitutive model parameters to the soil and rock mass, apply displacement boundary conditions and gravity loads, calculate the internal stress field of the slope using the finite element method, and save the stress components at the integration points of each element.

[0089] S2: Sliding surface parameterization and safety factor function construction: The circular arc sliding surface is uniquely determined by three parameters: the x-coordinate of the sliding point X1, the x-coordinate of the sliding point X2, and the sag Y. The circular arc sliding surface is discretized, and the normal stress σ of each micro-segment is calculated by combining finite element stress field interpolation. n Based on the tangential stress τ, and the cumulative anti-slip force and sliding force of the circular arc, a safety factor function Fs=f(X1,X2,Y) is constructed. By inputting three parameters, the corresponding safety factor is obtained. This safety factor function serves as the objective function for subsequent optimization search.

[0090] S3: Global Coarse Search - Artificial Intelligence Identifies Potential Local Extrema Regions: Within feasible limits, generate at least 500 sets of random sliding surface parameters based on the three parameters: the x-coordinate of the entry point (X1), the x-coordinate of the exit point (X2), and the sag (Y). Then, call the constructed safety factor function to calculate the safety factor F for each set of sliding surface parameters. {s,i} After normalizing the sliding surface parameter set, a clustering algorithm is used to identify multiple local extrema regions with low safety factors, and the parameter subspace of each local extrema region is determined, along with the initial point P used as the subsequent deterministic fine search. init .

[0091] S4: Local Fine-Search - Deterministic Method for Finding Local Minimum: Starting with an initial point P for each local minimum region. init Starting from the point P, a gradient-based deterministic optimization algorithm is used to iteratively search for the local minimum safety factor in the corresponding parameter subspace, and a dual convergence criterion and boundary projection mechanism are set up. The current point P_best_local and the corresponding safety factor F_min_local are obtained as the accurate minimum safety factor and optimal sliding surface parameters of the local region.

[0092] S5: Determine the global critical sliding surface: Compare the local minima and take the one with the smallest safety factor as the global critical sliding surface and its minimum safety factor.

[0093] S6: Output critical sliding surface results: Output minimum safety factor, sliding surface geometric parameters, discrete trajectory and visualization results.

[0094] In some embodiments, the finite element stress field calculation in step S1 includes:

[0095] S11. Geometric Modeling: Establish a two-dimensional or three-dimensional geometric model based on the slope's geometric dimensions (slope height, slope ratio, slope crest range, slope toe range). For two-dimensional analysis, the plane strain assumption is used; for three-dimensional analysis, a three-dimensional solid model of the slope is established.

[0096] S12. Material parameter assignment: Assign elastoplastic constitutive model parameters to the soil and rock mass, including: density ρ, elastic modulus E, Poisson's ratio ν, cohesion c, internal friction angle φ, and dilatation angle ψ, etc.

[0097] S13. Perform finite element mesh generation on the established geometric model. In the specific implementation process, quadrilateral (two-dimensional) or hexahedral (three-dimensional) elements are used, and local refinement is performed in the potential sliding area (near the slope surface and toe of the slope). The total number of mesh elements is controlled between 1000 and 5000.

[0098] S14. Boundary conditions and load application: Apply displacement boundary conditions: fix all degrees of freedom at the bottom and constrain horizontal displacement on both sides; apply gravity load (gravitational acceleration g) and consider possible groundwater pressure, seismic load, etc.

[0099] S15. Finite element solution: The Newton-Raphson iteration method is used for nonlinear elastoplastic solution to obtain the stress components at each element integration point. , , , , , The stress field results are saved as a data file for subsequent safety factor calculations.

[0100] Combined with appendix Figure 2 and attached Figure 3 In some embodiments, the sliding surface parameterization and safety factor function construction in step S2 includes:

[0101] S21. Determine the parameterization of the sliding surface: For a circular sliding surface, three parameters are used to uniquely determine it: the x-coordinate of the sliding point. (x-coordinate of the intersection of the circular sliding surface and the slope), x-coordinate of the exit point (The x-coordinate of the other intersection point of the circular sliding surface and the slope), the sag Y (range 0-Ymax, Ymax can be 0.5-1.5 times the slope height), such as Figure 2 As shown.

[0102] Coordinate system: X-axis horizontal to the right, Y-axis vertical downward (common in geotechnical engineering cross-sections);

[0103] Slope outline: The slope top, slope surface, and slope toe are formed by points ABCD and FE;

[0104] Circular arc sliding surface ( Figure 2 S in the diagram: An arc extending from the upper left (near point B) to the lower right (near point C);

[0105] in, The x-coordinate of the intersection of the sliding surface and the slope (the entry point); Y is the x-coordinate of the other intersection point of the sliding surface and the slope (the slip point); Y is the arc sag—that is, the vertical distance from the arc apex to the chord AB. Figure 2 (marked with a middle arrow).

[0106] S22. Establish geometric constraints: Calculate the coordinates of the circle center based on the three parameters X1, X2, and Y using geometric relationships. , The arc has a center on the perpendicular bisector of chord AB and a radius R. The arc's center lies on the perpendicular bisector of chord AB, and its sag Y is defined as the vertical distance from the apex of the arc to chord AB. It must be ensured that the arc lies entirely within the slope and does not extend beyond the boundary.

[0107] S23. Discretized Arc: Divide the arc into M equal-distance segments (M=100-200), and record the coordinates of the midpoint of each segment. , ) and the arc length of the micro-segment Δ ,like Figure 3 As shown;

[0108] S24. Stress field interpolation: For the coordinates of the midpoint of each micro-segment ( , The normal stress at that point is obtained from the finite element stress field saved in step S15 through interpolation (such as bilinear interpolation or shape function interpolation). and tangential stress τ.

[0109] S25. Calculate the safety factor. Accumulate the anti-slip force and the sliding force along the circular arc to obtain the safety factor. :

[0110] ;

[0111] Where c and φ are the cohesion and internal friction angle of the soil, respectively, and the cohesion and internal friction angle values ​​of the soil in the region where the sliding area integral point is located are selected; Here are the grid numbers that the sliding surface traverses; the total number of grids traversed is... indivual; For the number Normal stress at the integration point within a mesh element; For the number Tangential stress at the integration point within the mesh element; For the sliding surface in numbered Grid length;

[0112] S26: Construct a safety factor function; encapsulate the above calculation process into a function Fs=f( , The function takes three parameters (Y, y) as input and outputs the corresponding safety coefficient; this function serves as the objective function for subsequent optimization searches.

[0113] In some embodiments, the local coarse search-artificial intelligence identification of potential local extremum regions in step S3 includes:

[0114] S31. Determine the parameter sampling range: Based on the slope geometry, set the feasible domain boundaries for X1, X2, and Y. For example: X1∈[x_left,x_right], X2∈[x_left2,x_right2], Y∈[0,Ymax].

[0115] S32. Generate initial sample points: Within the feasible region, use Latin hypercube sampling or uniform random sampling to generate N sets of sliding surface parameters (N≥500), denoted as ( , , ), i=1,2,...,N.

[0116] S33. Calculate the safety factor for sample points: For each set of sliding surface parameters, call the safety factor function Fs=f( from step S26). , ,Y), calculate the corresponding safety factor F {s,i} .

[0117] S34. Constructing the dataset: Combine the three-dimensional parameters of the sample points and their corresponding safety factors into four-dimensional data points: ( , , F {s,i} ).

[0118] S35. Data normalization: Normalize each dimension of the four-dimensional data (e.g., Min-Max normalization to the [0,1] interval) to eliminate the influence of dimensions and facilitate clustering.

[0119] S36. Cluster Analysis: Density-based clustering algorithms (such as DBSCAN) or Gaussian Mixture Models (GMM) are used to cluster the normalized dataset. The parameter space (X1, X2, Y) is the primary feature used during clustering, with the safety factor serving as an auxiliary feature.

[0120] S37. Extract Clusters: Extract clusters with low safety factors (i.e., the mean Fs within a cluster is less than the overall mean) from the clustering results. Each cluster represents a potential local extremum region. If DBSCAN identifies noise points (unclassified points), discard them.

[0121] S38. Determine the parameter subspace for each local extremum region (i.e., each cluster). For each cluster, calculate the minimum and maximum values ​​of its parameters to form a three-dimensional rectangular subspace: [ , ]×[ , ]×[ , ].

[0122] S39. Select the initial point for the local search. Within each cluster, select the sample point with the lowest safety factor, or select the cluster center point (mean of each dimension), as the initial point P for the subsequent deterministic fine search. init .

[0123] In some embodiments, the local fine-search-deterministic method for finding local minima in step S4 includes:

[0124] S401. Set local search parameters: Define the maximum number of iterations Tmax (e.g., 50 iterations), convergence tolerance. (Gradient magnitude threshold, such as 10) -3 )and (Parameter change threshold, such as 10) -4 ).

[0125] S402. Initialize the current point: Let the current parameter vector P... curr =P init .

[0126] S403. Calculate the safety factor of the current point: Call the function f(P) curr ) to obtain F curr .

[0127] S404. Calculate the gradient of the objective function: The gradient ∇F(P) is approximated using the central difference method. curr ):

[0128] ;

[0129] The step size coefficient for the sliding point can be taken as the sliding point. 0.01-0.05 times the range of variation;

[0130] ;

[0131] The step size coefficient for the slip-out point can be taken as the slip-out point. 0.01-0.05 times the range of variation;

[0132] ;

[0133] The step size coefficient for the sag can be taken as the sag. 0.01-0.05 times the range of variation;

[0134] S405. Check convergence condition: If ||∇F|| < Or the parameter changes in two consecutive iterations < If the convergence occurs, proceed to step S412.

[0135] S406. Determine the search direction. Use the steepest descent method, with the search direction d = -∇F (negative gradient direction). The conjugate gradient method or the BFGS method can be used to update the direction to accelerate convergence.

[0136] S407. One-dimensional line search for optimal step size. In direction d, find the step size α× such that F(P) curr Minimize +α·d). This can be achieved using the golden section search or Armijo backtracking.

[0137] S408. Update parameters. P new =P curr +α×·d.

[0138] S409. Boundary constraint handling. If P new If the parameter subspace boundary of the local region determined in step S38 is exceeded, then P will be... new Project back onto the boundary (take the boundary value).

[0139] S410. Update the iteration count. t = t + 1, and force a stop if t > Tmax.

[0140] S411. Repeat the iteration. Let P curr =Pnew Return to step S403.

[0141] S412. Record the local optimal solution. When convergence occurs, record the current point P_best_local and the corresponding safety factor F_min_local, which serve as the precise minimum safety factor and optimal sliding surface parameters for that local region.

[0142] In some embodiments, determining the global critical sliding surface in step S5 includes:

[0143] S51. Collect the optimal solutions for each local region. Summarize the (P_best_local, F_min_local) of each local extreme region obtained in step S412 into a list.

[0144] S52. Compare global minimum values. Iterate through the list, find the entry with the smallest safety factor, and record its parameters P_global and F_min_global.

[0145] S53. Global Minimum Verification (Optional). Optionally, a small range of random sampling (e.g., 10 points) can be performed near P_global to verify that the point is indeed the global minimum, preventing omissions due to numerical errors.

[0146] In some embodiments, the output of the critical sliding surface result in step S6 includes:

[0147] S61: Output the minimum safety factor. Output F_min_global, rounded to two decimal places.

[0148] S62: Outputs the geometric parameters of the critical sliding surface. Outputs the x-coordinate of the sliding point (X1), the x-coordinate of the sliding point (X2), the elevation (Y), and the calculated coordinates of the center of the circle. , ) and radius R.

[0149] S63: Output the sliding surface trajectory. Generate a discrete point coordinate sequence of the arc based on the parameters. , It can be exported as a text file or DXF format.

[0150] S64: Visualization. Draw the critical slip surface on the slope profile and label the safety factor value. You can also draw other local extreme slip surfaces (such as dashed lines).

[0151] S65: Generate Search Report. Automatically generate a search process report, including: total number of calculations, convergence status of each local region, iteration history curves, and final results.

[0152] Example 1

[0153] Implementation scenario: A homogeneous clay slope with a height of 20m and a slope ratio of 1:1.5. Soil parameters: density ρ = 1.9g / cm³. 3 Given a slope with cohesion c = 25 kPa, internal friction angle φ = 18°, elastic modulus E = 30 MPa, and Poisson's ratio ν = 0.35, we need to search for the critical circular sliding surface and its minimum safety factor.

[0154] (1) Implementation details of step S1:

[0155] A two-dimensional plane strain finite element model is established. The slope range is taken as 30m to the left of the slope toe and 40m to the right of the slope top. The bottom boundary depth is taken as twice the slope height (40m).

[0156] The Mohr-Coulomb elastoplastic constitutive model was used to divide the data into quadrilateral elements, totaling approximately 2000 elements.

[0157] Applying a self-weight load (gravitational acceleration g = 9.8 m / s²) 2 ), constraining the fixed displacement at the bottom and the horizontal constraints on both sides.

[0158] By performing a nonlinear finite element method, the stress components at each integration point are obtained. , , , , , Save the stress results as a mesh file.

[0159] (2) Implementation details of step S2:

[0160] Define the sliding surface parameters: the x-coordinate of the sliding point X1 (range: 0-10m horizontally to the left of the slope toe), the x-coordinate of the sliding point X2 (range: 0-15m horizontally to the right of the slope crest), and the sag Y (range: 0-15m).

[0161] Given a set of parameters (X1, X2, Y), the arc is determined by geometric construction: with the entry point A and the exit point B as chords, the arch height is set as Y, and the position of the center O and the radius R are calculated.

[0162] Discretize the arc into 100 micro-segments, and the coordinates of the midpoint of each micro-segment are ( , ).

[0163] The normal stress σ_n and tangential stress τ at this point can be obtained by bilinear interpolation from the finite element stress field.

[0164] Calculate the safety factor Fs=Σ(c+σ_n·tanφ)·Δl / Σ(τ·Δl).

[0165] (3) Implementation details of step S3:

[0166] Within the feasible region of parameters, 1000 sets of random parameters (X1, X2, Y) are generated using Latin hypercube sampling.

[0167] Calculate the safety factor Fs for each set of parameters.

[0168] Input 1000 sample points (3D parameters + 1D safety factor) into the DBSCAN clustering algorithm, set the clustering radius ε=0.15 (normalized parameter space), and the minimum number of samples MinPts=10.

[0169] DBSCAN clustering results identified three potential minimum clusters (i.e., local extreme value regions):

[0170] Region 1: X1≈[-2,0]m, X2≈[8,12]m, Y≈[2,4]m, corresponding to shallow slip;

[0171] Region 2: X1≈[0,3]m, X2≈[12,16]m, Y≈[6,9]m, corresponding to the middle layer sliding;

[0172] Region 3: X1≈[4,7]m, X2≈[18,22]m, Y≈[10,13]m, corresponding to deep sliding.

[0173] Record the center point of each region as the initial point for subsequent deterministic searches.

[0174] (4) Implementation details of step S4:

[0175] For region 1, with the center point P1=(X1=-1m,X2=10m,Y=3m) as the initial point, the steepest descent method is used for iteration:

[0176] Calculate the safety factor Fs and gradient ∇Fs at the current point (approximate calculation using the central difference method, with a step size Δ=0.05m).

[0177] Perform a one-dimensional line search along the negative gradient direction to find the optimal step size α.

[0178] Update parameter: Xnew=Xold-α·∇Fs.

[0179] Repeat the iteration until the gradient magnitude is less than 1e-3 or the parameter change is less than 1e-4.

[0180] After 8 iterations, the system converges, yielding a local optimum P1×=(X1=-1.2m,X2=10.5m,Y=3.2m) with a minimum safety factor Fs1=1.15.

[0181] For region 2, with P2=(X1=1.5m,X2=14m,Y=7.5m) as the initial point, the steepest descent method is also used. After 12 iterations, it converges to obtain P2×=(X1=1.8m,X2=14.8m,Y=8.1m) and Fs2=1.08.

[0182] For region 3, with P3=(X1=5.5m,X2=20m,Y=11.5m) as the initial point, it converges after 15 iterations, resulting in P3×=(X1=6.2m,X2=21.3m,Y=12.4m) and Fs3=0.96.

[0183] (5) Implementation details of step S5:

[0184] Comparing the three local minimum safety factors: Fs1=1.15, Fs2=1.08, and Fs3=0.96, the minimum value is Fs3=0.96, corresponding to the deep sliding surface in region 3, with parameters (X1=6.2m, X2=21.3m, Y=12.4m). This sliding surface is the critical sliding surface of the slope, and the safety factor 0.96 < 1.0 indicates that the slope is in an unstable state and requires reinforcement.

[0185] (6) Implementation details of step S6:

[0186] Output global minimum safety factor: 0.96

[0187] Output critical sliding surface parameters: X1 = 6.2m (right side of the slope toe), X2 = 21.3m (right side of the slope crest), and Y = 12.4m (slip point x-coordinate).

[0188] Draw the circular sliding surface on the slope profile and mark the safety factor.

[0189] Search report generation: Total calculations: 1045 (1000 coarse searches + approximately 15 fine searches each), taking approximately 30 seconds.

[0190] The method of this embodiment is compared with the prior art to verify the results:

[0191] Using only the traditional genetic algorithm (population 100, generation 200), it is necessary to calculate the sliding surface 20,000 times, which takes about 120 seconds, and the minimum safety factor found is 0.958.

[0192] Using only the steepest descent method (randomly selected initial point), if the initial point is selected in region 1, it converges to F_s=1.15, and the error judgment is stable; only by selecting region 3 can 0.96 be found, but the success rate of randomly selecting the initial point is low (about 1 / 3).

[0193] The method of this invention requires only about 1045 calculations of the sliding surface, taking about 30 seconds, and guarantees finding the global minimum value of 0.96. Compared with the genetic algorithm, the computational cost is reduced by about 48%; compared with the steepest descent method, the success rate reaches 100% (avoiding initial point dependence).

[0194] It should be understood that the above description of the preferred embodiments is quite detailed, but it should not be considered as a limitation on the scope of protection of this invention. Those skilled in the art, under the guidance of this invention, can make substitutions or modifications without departing from the scope of protection of the claims of this invention, and all such substitutions or modifications fall within the scope of protection of this invention. The scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for searching the critical sliding surface of a slope based on artificial intelligence, characterized in that, include: S1: Finite element stress field calculation: Establish the slope geometric model, assign elastoplastic constitutive model parameters to the soil and rock mass, apply displacement boundary conditions and gravity loads, use the finite element method to calculate the internal stress field of the slope, and save the stress components of each element integration point. S2: Sliding surface parameterization and safety factor function construction: The circular arc sliding surface is uniquely determined by three parameters: the x-coordinate of the sliding point X1, the x-coordinate of the sliding point X2, and the sag Y. The circular arc sliding surface is discretized, and the normal stress σ of each micro-segment is calculated by combining finite element stress field interpolation. n Based on the tangential stress τ, and the cumulative anti-slip force and sliding force of the circular arc, a safety factor function Fs=f(X1,X2,Y) is constructed. By inputting three parameters, the corresponding safety factor is obtained. This safety factor function serves as the objective function for subsequent optimization search. S3: Global Coarse Search - Artificial Intelligence Identifies Potential Local Extrema Regions: Within feasible limits, generate at least 500 sets of random sliding surface parameters based on the three parameters: the x-coordinate of the entry point (X1), the x-coordinate of the exit point (X2), and the sag (Y). Then, call the constructed safety factor function to calculate the safety factor F for each set of sliding surface parameters. {s,i} After normalizing the sliding surface parameter set, a clustering algorithm is used to identify multiple local extrema regions with low safety factors, and the parameter subspace of each local extrema region is determined, along with the initial point P used as the subsequent deterministic fine search. init ; S4: Local Fine-Search - Deterministic Method for Finding Local Minimum: Starting with an initial point P for each local minimum region. init Starting from point P, a gradient-based deterministic optimization algorithm is used to iteratively search for the local minimum safety factor in the corresponding parameter subspace, and a dual convergence criterion and boundary projection mechanism are set up. The current point P_best_local and the corresponding safety factor F_min_local are obtained as the accurate minimum safety factor and optimal sliding surface parameters of the local region. S5: Determine the global critical sliding surface: Compare the local minima and take the one with the smallest safety factor as the global critical sliding surface and its minimum safety factor; S6: Output critical sliding surface results: Output minimum safety factor, sliding surface geometric parameters, discrete trajectory and visualization results.

2. The artificial intelligence-based slope critical sliding surface search method according to claim 1, characterized in that, The finite element stress field calculation in step S1 includes: S11. Geometric Modeling: Establish a two-dimensional or three-dimensional geometric model based on the geometric dimensions of the slope; S12. Assigning material parameters: Assigning elastic-plastic constitutive model parameters to the rock and soil mass, including density ρ, elastic modulus E, Poisson's ratio ν, cohesion c, internal friction angle φ, and dilatation angle ψ; S13. Perform finite element mesh generation on the established geometric model; S14. Boundary Conditions and Load Application: Apply displacement boundary conditions: fix all degrees of freedom at the bottom, constrain horizontal displacement on both sides; apply gravity load. S15. Finite element solution: The Newton-Raphson iteration method is used for nonlinear elastoplastic solution to obtain the stress components at each element integration point. , , , , , The stress field results are saved as a data file for subsequent safety factor calculations.

3. The artificial intelligence-based slope critical sliding surface search method according to claim 1, characterized in that, The sliding surface parameterization and safety factor function construction in step S2 include: S21. Determine the parameterization of the sliding surface: For a circular sliding surface, three parameters are used to uniquely determine it: the x-coordinate of the sliding point. x-coordinate of the sliding point 、Ya Taka; S22. Establish geometric constraints: Calculate the coordinates of the circle center based on the three parameters X1, X2, and Y using geometric relationships. , and radius R; S23. Discretized Circular Arc: Divide the circular arc of the sliding surface into M equal segments, and record the coordinates of the midpoint of each segment. , ) and the arc length of the micro-segment Δ ; S24. Stress field interpolation: For the coordinates of the midpoint of each micro-segment ( , The normal stress at that point is obtained by interpolation from the finite element stress field saved in step S15. and tangential stress τ; S25. Calculate the safety factor; accumulate the anti-slip force and sliding force along the arc to obtain the safety factor. : ; Where c and φ are the cohesion and internal friction angle of the soil, respectively; Here are the grid numbers that the sliding surface traverses; the total number of grids traversed is... indivual; For the number Normal stress at the integration point within a mesh element; For the number Tangential stress at the integration point within the mesh element; For the sliding surface in the numbered Grid length; S26: Construct a safety factor function; encapsulate the above calculation process into a function Fs=f( , The function takes three parameters (Y, y) as input and outputs the corresponding safety coefficient; this function serves as the objective function for subsequent optimization searches.

4. The artificial intelligence-based slope critical sliding surface search method according to claim 3, characterized in that, The local coarse search-artificial intelligence identification of potential local extremum regions in step S3 includes: S31. Determine the parameter sampling range: Based on the slope geometry, set the feasible domain boundaries for X1, X2, and Y; S32. Generate initial sample points: Within the feasible region, use Latin hypercube sampling or uniform random sampling to generate N sets of sliding surface parameters, where N≥500, denoted as ( , , ), i=1,2,...,N; S33. Calculate the safety factor for sample points: For each set of sliding surface parameters, call the safety factor function Fs=f( from step S26). , ,Y), calculate the corresponding safety factor F {s,i} ; S34. Constructing the dataset: Combine the three-dimensional parameters of the sample points and their corresponding safety factors into four-dimensional data points: ( , , F {s,i} ); S35. Data Normalization: Normalize each dimension of the four-dimensional data; S36. Cluster Analysis: Cluster the normalized dataset using a density-based clustering algorithm; S37. Extract clusters: Extract clusters with low safety factors from the clustering results. Each cluster represents a potential local extremum region. S38. Determine the parameter subspace for each local extremum region; for each cluster, calculate the minimum and maximum values ​​of its parameters to form a three-dimensional rectangular subspace: [ , ]×[ , ]×[ , ]; S39. Select the initial point for the local search; within each cluster, select the sample point with the lowest safety factor, or select the cluster center point (mean of each dimension), as the initial point P for the subsequent deterministic fine search. init .

5. The artificial intelligence-based slope critical sliding surface search method according to claim 4, characterized in that, The local fine-search-deterministic method for finding local minima in step S4 includes: S401. Set local search parameters: Define the maximum number of iterations Tmax and convergence tolerance. and ; S402. Initialize the current point: Let the current parameter vector P... curr =P init ; S403. Calculate the safety factor of the current point: Call the function f(P) curr ) to obtain F curr ; S404. Calculate the gradient of the objective function: The gradient ∇F(P) is approximated using the central difference method. curr ): ; The step size coefficient for the sliding point can be taken as the sliding point. 0.01-0.05 times the range of variation; ; The step size coefficient for the slip-out point can be taken as the slip-out point. 0.01-0.05 times the range of variation; ; The step size coefficient for the sag can be taken as the sag. 0.01-0.05 times the range of variation; S405. Check convergence condition: If ||∇F|| < Or the parameter changes in two consecutive iterations < If the convergence occurs, proceed to step S412. S406. Determine the search direction: Use the steepest descent method, and the search direction is d = -∇F (negative gradient direction); S407. One-dimensional line search for optimal step size: In direction d, solve for the step size α× such that F(P) curr Minimize +α·d); S408. Update parameters: P new =P curr +α×·d; S409. Boundary Constraint Handling: If P new If the parameter subspace boundary of the local region determined in step S38 is exceeded, then P will be... new Project back onto the boundary; S410. Update iteration count: t = t + 1, if t > Tmax, then force stop; S411. Repeated Iteration: Let P curr =P new Return to step S403; S412. Record the local optimal solution: When convergence occurs, record the current point P_best_local and the corresponding safety factor F_min_local as the precise minimum safety factor and optimal sliding surface parameters for the local region.

6. The artificial intelligence-based slope critical sliding surface search method according to claim 5, characterized in that, Determining the global critical sliding surface in step S5 includes: S51. Collect the optimal solutions for each local region: Summarize the (P_best_local, F_min_local) of each local extreme value region obtained in step S412 into a list: S52. Compare global minimum values. Iterate through the list, find the entry with the smallest safety factor, and record its parameters P_global and F_min_global.

7. The artificial intelligence-based slope critical sliding surface search method according to claim 6, characterized in that, The critical sliding surface result output in step S6 includes: S61: Output minimum safety factor: Output F_min_global; S62: Output critical sliding surface geometric parameters: Output the x-coordinate of the sliding point X1, the x-coordinate of the sliding point X2, and the sag Y, as well as the calculated center coordinates ( , and radius R; S63: Output sliding surface trajectory: Generate a discrete point coordinate sequence of the arc based on the parameters ( , Export as a text file or DXF format; S64: Visualization: Draw the critical sliding surface on the slope profile and label the safety factor value.

8. The artificial intelligence-based slope critical sliding surface search method according to claim 7, characterized in that, The output of the critical sliding surface result in step S6 also includes: Generate a search report: Automatically generate a search process report, including the total number of calculations, convergence status of each local region, iteration history curves, and final results.