A bank landslide soil body shear strength parameter inversion method, device, equipment and storage medium

Abstract

The present application relates to the technical field of data processing, and provides a reservoir bank landslide soil body shear strength parameter inversion method, device, equipment and storage medium, the effective value range of soil body shear strength parameter is determined by acquiring target safety factor, and the training sample set is generated in the effective value range through numerical simulation, the safety factor prediction model is constructed through machine learning method according to the training sample set, and the safety factor prediction model and target safety factor are searched in the effective value range, and the soil body shear strength parameter inversion result matched with target safety factor is determined.The present application realizes the accurate acquisition of the reservoir bank landslide soil body shear strength parameter, releases more data value by deep mining existing information in the survey stage, realizes the automation and high efficiency of the inversion process, completely gets rid of the dependence on artificial experience trial, and finally provides scientific basis for the engineering design optimization and disaster prevention and control decision of reservoir bank landslide.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, equipment, and storage medium for inverting shear strength parameters of landslide soil on reservoir banks. Background Technology

[0002] Currently, obtaining shear strength parameters of landslide soil on reservoir banks mainly relies on traditional exploration methods, such as in-situ tests and laboratory geotechnical tests. While these methods can provide basic data, they have significant limitations: on the one hand, due to limitations in exploration costs, site conditions, and testing techniques, the sample size of parameters obtained is limited, making it difficult to comprehensively reflect the parameter differences in different areas of the landslide body; on the other hand, the exploration phase accumulates a large amount of geological exploration data and engineering geological information, but existing technologies have failed to fully explore the potential correlations of this data, resulting in low information utilization and an inability to provide richer support for parameter inversion.

[0003] Furthermore, even when using a combination of numerical simulation and inverse analysis, the traditional inversion process heavily relies on engineers' experience for repeated trial and error. Each parameter adjustment requires rerunning the time-consuming finite element calculation, resulting in extremely low inversion efficiency and making it difficult to meet the needs of engineering design and rapid evaluation.

[0004] Therefore, there is an urgent need for a method that integrates the advantages of computer technology, can deeply mine survey information, and accurately obtain the shear strength parameters of reservoir bank landslide soil, so as to meet the actual needs of engineering design optimization and efficient disaster prevention and control. Summary of the Invention

[0005] The main objective of this invention is to provide a method, apparatus, equipment, and storage medium for inverting shear strength parameters of landslide soil on reservoir banks, in order to solve the technical problem of extremely low inversion efficiency in obtaining shear strength parameters of landslide soil on reservoir banks, due to the limited and unreasonable distribution of traditional survey samples leading to insufficient data representativeness and the lack of in-depth exploration of the value of existing survey and simulation data.

[0006] To achieve the above objectives, this invention provides a method for inverting the shear strength parameters of soil in reservoir bank landslides, comprising the following steps:

[0007] S1, obtain the target safety factor;

[0008] S2, determine the effective range of soil shear strength parameters, and generate a training sample set within the effective range through numerical simulation; wherein, the training sample set includes multiple sets of soil shear strength parameters and corresponding safety factors.

[0009] S3, Based on the training sample set, a safety factor prediction model is constructed using machine learning methods; wherein, the safety factor prediction model includes the mapping relationship between soil shear strength parameters and safety factors;

[0010] S4. Based on the safety factor prediction model and the target safety factor, a search is performed within the effective value range to determine the soil shear strength parameter inversion result that matches the target safety factor.

[0011] Preferably, step S2 includes the following steps:

[0012] S21. N sets of initial soil shear strength parameter sample points are uniformly generated within the effective value range. For each set of initial soil shear strength parameter sample points, slope stability is simulated by finite element numerical simulation. The corresponding initial safety factor is output. Then, all initial soil shear strength parameter sample points and corresponding initial safety factors are collected to obtain the initial training sample set.

[0013] S22, using the initial training sample set as the current training sample set, perform K rounds of iterative sampling; wherein each round of iteration includes steps S221~S223:

[0014] S221, Based on all soil shear strength parameter sample points in the current training sample set, construct Thiessen polygons within the effective value range, and divide the effective value range into multiple Thiessen polygon units; wherein, each Thiessen polygon unit contains one soil shear strength parameter sample point.

[0015] S222, randomly generate M candidate soil shear strength parameter sample points within the effective value range; wherein, for each candidate soil shear strength parameter sample point, add it to the current training sample set to form a temporary sample set; construct a temporary safety factor prediction proxy model based on the temporary sample set, and use the cross-validation method to calculate the prediction error of the temporary safety factor prediction proxy model, and then record the prediction error as the evaluation error of the candidate soil shear strength parameter sample point;

[0016] S223, from the M candidate soil shear strength parameter sample points, select the candidate soil shear strength parameter sample point with the largest evaluation error as the optimal supplementary sample point for this round; perform finite element numerical simulation of slope stability on the optimal supplementary sample point for this round, and output the corresponding supplementary safety factor; then add the optimal supplementary sample point for this round and the corresponding supplementary safety factor to the current training sample set, and update the current training sample set;

[0017] S23, after completing K rounds of iterative sampling, the final updated current training sample set is used as the training sample set.

[0018] Preferably, step S222, which involves constructing a temporary security coefficient prediction proxy model based on the temporary sample set, includes the following steps:

[0019] The soil shear strength parameters in the temporary sample set are standardized to obtain a standardized temporary sample set.

[0020] The Support Vector Regression (SVR) algorithm is used, and a set of preset hyperparameters are configured for the SVR algorithm. The soil shear strength parameters in the standardized temporary sample set are used as input, and the corresponding safety factor in the standardized temporary sample set is used as output to train the temporary safety factor prediction surrogate model.

[0021] Preferably, the step S222, which uses cross-validation to calculate the prediction error of the temporary safety factor prediction proxy model, specifically includes the following steps:

[0022] The standardized temporary sample set is randomly divided into P mutually exclusive subsets;

[0023] Perform P rounds of verification sequentially; in each round of verification, perform the following steps:

[0024] Select one subset from the P subsets as the validation set for the current round, and combine the remaining P-1 subsets as the training set for the current round;

[0025] Using the preset hyperparameters, the support vector regression model is retrained with the current round training set to obtain the support vector regression model, and the support vector regression model is used as the validation model for the current round.

[0026] The current round's verification model is used to predict the current round's verification set to obtain the predicted safety coefficient;

[0027] The mean square error between the predicted safety factor and the actual safety factor inherent in the current round of verification is calculated and used as the verification error of the current round.

[0028] Calculate the average of all P verification errors obtained in the P rounds of verification, and use the average as the prediction error of the temporary safety factor prediction surrogate model.

[0029] Preferably, step S3 includes the following steps:

[0030] S31, Standardize the soil shear strength parameters in the training sample set to obtain a standardized training sample set;

[0031] S32, based on the standardized training sample set, a sparrow search algorithm is used to globally optimize the hyperparameters of the support vector regression model; wherein, the sparrow search algorithm is executed in each iteration: using the hyperparameter combination of the current iteration, a support vector regression model is constructed based on the standardized training sample set and cross-validated, and the calculated mean squared error is used as the fitness value of the hyperparameter combination of the current iteration; the sparrow search algorithm updates iteratively, searching for the hyperparameter combination that minimizes the fitness value as the optimal hyperparameter combination;

[0032] S33. Using the optimal hyperparameter combination and the standardized training sample set, the support vector regression model is trained to obtain the optimized safety coefficient prediction model.

[0033] Preferably, step S4 includes the following steps:

[0034] S41, within the effective value range, the cohesion and internal friction angle are discretized at equal intervals to generate a sequence of cohesion parameter values ​​and a sequence of internal friction angle parameter values.

[0035] S42, combine the cohesion parameter value sequence and the internal friction angle parameter value sequence in pairs to generate a two-dimensional grid point set covering the effective value range;

[0036] S43, input the cohesion and internal friction angle corresponding to each grid point in the two-dimensional grid point set into the safety factor prediction model; calculate the input cohesion and internal friction angle through the safety factor prediction model, output the predicted safety factor corresponding to each grid point, and obtain a set of predicted safety factors corresponding to each grid point.

[0037] S44, calculate the absolute error between each predicted safety factor in the predicted safety factor set and the target safety factor; find the grid point with the smallest absolute error from the two-dimensional grid point set, and use the cohesion and internal friction angle corresponding to the grid point as the soil shear strength parameter inversion result that matches the target safety factor.

[0038] Preferably, the number of initial soil shear strength parameter sample points N ranges from 5 to 20, the number of iteration sampling rounds K ranges from 50 to 200, and the value of M ranges from 800 to 1200.

[0039] The present invention also provides a device for inverting the shear strength parameters of landslide soil on reservoir banks, used to perform the inversion method for the shear strength parameters of landslide soil on reservoir banks as described above, including:

[0040] The target input module is used to obtain the target safety factor.

[0041] The sample generation module is used to determine the effective range of soil shear strength parameters and generate a training sample set through numerical simulation within the effective range; wherein, the training sample set includes multiple sets of soil shear strength parameters and corresponding safety factors.

[0042] The model building module is used to build a safety factor prediction model based on the training sample set using machine learning methods; wherein, the safety factor prediction model includes the mapping relationship between soil shear strength parameters and safety factors;

[0043] The inversion search module is used to search within the effective value range based on the safety factor prediction model and the target safety factor to determine the soil shear strength parameter inversion result that matches the target safety factor.

[0044] The present invention also provides a device for inverting the shear strength parameters of soil in a reservoir bank landslide, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described method for inverting the shear strength parameters of soil in a reservoir bank landslide.

[0045] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for inverting the shear strength parameters of a reservoir bank landslide soil.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] This invention provides a method, apparatus, equipment, and storage medium for inverting shear strength parameters of soil in reservoir bank landslides. This application achieves accurate acquisition of shear strength parameters of soil in reservoir bank landslides, and simultaneously releases more data value by deeply mining existing information from the exploration phase, ultimately providing a scientific basis for engineering design optimization and disaster prevention decisions for reservoir bank landslides. Specifically, by determining the effective value range and generating a training sample set through numerical simulation, it overcomes the sample sparsity caused by cost and condition limitations in traditional exploration; by constructing a safety factor prediction model using machine learning methods, it can deeply mine and learn the complex, nonlinear intrinsic laws between soil shear strength parameters and safety factors from limited sample data, fully releasing the potential value of previous exploration and simulation data; based on the safety factor prediction model and the target safety factor, it searches within the effective value range to determine matching inversion results, realizing the automation and efficiency of the inversion process, completely eliminating the dependence on manual experience-based calculations. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0049] Figure 1 This is a schematic flowchart of one embodiment of the present invention;

[0050] Figure 2 This is a spatial scatter plot illustrating the soil shear strength parameters in one embodiment of the present invention.

[0051] Figure 3 This is a schematic diagram of a two-dimensional planar model of a shallow landslide in an early warning zone, created using slope stability analysis software, according to one embodiment of the present invention.

[0052] Figure 4 This is a schematic diagram illustrating the relationship between the number of samples and the average relative error in one embodiment of the present invention.

[0053] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0054] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0056] In this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Furthermore, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0057] Please refer to Figures 1 to 4 The present invention provides a method for inverting the shear strength parameters of soil in a reservoir bank landslide, comprising the following steps:

[0058] S1, Obtain the target safety factor; wherein, the target safety factor is an input to the engineering design, which may be derived from national or industry-issued slope engineering design specifications, or a design safety standard for a specific treatment project that exceeds the minimum requirements of the specifications. For example, the target safety factor may be set to 1.15;

[0059] S2, determine the effective range of soil shear strength parameters, and generate a training sample set within the effective range through numerical simulation; wherein, the training sample set includes multiple sets of soil shear strength parameters and corresponding safety factors; the effective range in this step can be determined based on engineering geological survey data (such as drilling, sampling, and test results) to determine the parameters to be inverted (such as cohesion c and internal friction angle). The effective range of values ​​for (cohesion c, internal friction angle) is then determined through numerical simulation within this effective range. The training sample set of data pairs (safety factor Fs). For example, the soil shear strength parameters and corresponding safety factors for each group are (cohesion c, internal friction angle Fs). Safety factor Fs);

[0060] S3, Based on the training sample set, a safety factor prediction model is constructed using machine learning methods; wherein, the safety factor prediction model includes the mapping relationship between soil shear strength parameters and safety factors;

[0061] S4. Based on the safety factor prediction model and the target safety factor, a search is performed within the effective value range to determine the soil shear strength parameter inversion result that matches the target safety factor.

[0062] This application achieves accurate acquisition of shear strength parameters of soil in reservoir bank landslides. Simultaneously, it unlocks greater data value by deeply mining existing information from the exploration phase, ultimately providing a scientific basis for engineering design optimization and disaster prevention decisions in reservoir bank landslides. Specifically, by determining the effective value range and generating a training sample set through numerical simulation, it overcomes the sample sparsity limitations caused by cost and conditions in traditional exploration. By constructing a safety factor prediction model using machine learning methods, it can deeply mine and learn the complex, nonlinear intrinsic laws between soil shear strength parameters and safety factors from limited sample data, fully releasing the potential value of previous exploration and simulation data. Based on the safety factor prediction model and the target safety factor, a search is performed within the effective value range to determine matching inversion results, achieving automation and efficiency in the inversion process and completely eliminating reliance on manual empirical calculations.

[0063] In a preferred embodiment, step S2 includes the following steps:

[0064] S21. N sets of initial soil shear strength parameter sample points are uniformly generated within the effective value range. For each set of initial soil shear strength parameter sample points, slope stability is simulated by finite element numerical simulation. The corresponding initial safety factor is output. Then, all initial soil shear strength parameter sample points and corresponding initial safety factors are collected to obtain the initial training sample set.

[0065] Within the predetermined effective range of soil shear strength parameters (such as cohesion and internal friction angle), this step generates N sets of initial soil shear strength parameter sample points using a uniform sampling method. These serve as the foundational data for constructing the surrogate model. The purpose of uniform sampling is to cover the parameter space as broadly as possible without any prior information, providing a basic spatial framework for subsequent iterations. For each set of initial soil shear strength parameter sample points, a mature finite element method (FEM) is used to perform finite element numerical simulation of slope stability (e.g., Slide slope stability analysis software) to obtain the corresponding safety factor. The data from these N sets of initial soil shear strength parameter sample points are then aggregated to form an initial training sample set, which serves as the starting point for subsequent sampling. As a preferred example, the number N of initial soil shear strength parameter sample points ranges from 5 to 20.

[0066] S22, using the initial training sample set as the current training sample set, perform K rounds of iterative sampling; wherein, each round of iteration includes steps S221~S223: as a preferred example, the value range of the round K of iterative sampling is 50-200.

[0067] S221, Based on all soil shear strength parameter sample points in the current training sample set, construct Thiessen polygons within the effective value range, and divide the effective value range into multiple Thiessen polygon units; wherein, each Thiessen polygon unit contains one soil shear strength parameter sample point; this step constructs Thiessen polygons within the effective value range based on the coordinates (i.e., the values ​​of cohesion and internal friction angle) of all soil shear strength parameter sample points in the current training sample set, and divides the entire effective value range into multiple non-overlapping convex polygon units, each Thiessen polygon unit containing an existing soil shear strength parameter sample point, thereby intuitively defining the influence range of each sample point in the parameter space.

[0068] S222, randomly generate M candidate soil shear strength parameter sample points within the effective value range; wherein, for each candidate soil shear strength parameter sample point, add it to the current training sample set to form a temporary sample set; construct a temporary safety factor prediction proxy model based on the temporary sample set, and use the cross-validation method to calculate the prediction error of the temporary safety factor prediction proxy model, and then record the prediction error as the evaluation error of the candidate soil shear strength parameter sample point;

[0069] To identify the areas most in need of supplementary samples, M candidate soil shear strength parameter sample points are randomly generated within the effective value range to ensure a thorough exploration of the parameter space. As a preferred example, the value of M ranges from 800 to 1200. For each candidate soil shear strength parameter sample point, it is added to the current training sample set to form a temporary sample set. Based on the temporary sample set, a support vector regression model is quickly trained using a set of preset fixed hyperparameters (e.g., the support vector regression model penalty coefficient and kernel parameters set empirically), serving as a temporary safety factor prediction surrogate model. The purpose of using fixed hyperparameters is to ensure that all candidate soil shear strength parameter sample points are evaluated under the same model configuration benchmark, avoiding evaluation fluctuations introduced by hyperparameter differences and ensuring the fairness and comparability of the evaluation. P-fold cross-validation (e.g., P=5) is applied to the temporary safety factor prediction surrogate model. This step dynamically diagnoses the areas where the prediction accuracy of the temporary safety factor prediction surrogate model is weakest (i.e., the areas with the largest evaluation errors) under the current sample distribution, thus providing a clear direction for sampling.

[0070] S223, from the M candidate soil shear strength parameter sample points, select the candidate soil shear strength parameter sample point with the largest evaluation error as the optimal supplementary sample point for this round; perform finite element numerical simulation of slope stability on the optimal supplementary sample point for this round, and output the corresponding supplementary safety factor; then add the optimal supplementary sample point for this round and the corresponding supplementary safety factor to the current training sample set, update the current training sample set, thereby completing the expansion and update of the sample for this round of iteration.

[0071] S23, after completing K rounds of iterative sampling, the final updated current training sample set is used as the training sample set.

[0072] This embodiment generates a training sample set with optimized sample distribution using a limited number of efficient numerical simulations. Compared to traditional sampling, this embodiment guides sample points to automatically cluster in key regions with high prediction uncertainty. Therefore, with the same or even fewer total samples, the training sample set obtained in this embodiment can support the construction of a surrogate model with higher global prediction accuracy and stronger robustness. This particularly ensures that the surrogate model maintains high prediction reliability in parameter regions where data is relatively sparse but inversion search may reach.

[0073] In a preferred embodiment, step S222, which involves constructing a temporary security factor prediction proxy model based on the temporary sample set, includes the following steps:

[0074] The soil shear strength parameters in the temporary sample set are standardized to obtain a standardized temporary sample set. The standardization process can be the Z-score standardization method, which is a well-known and mature method and will not be elaborated here.

[0075] A Support Vector Regression (SVR) algorithm is employed, with a set of preset hyperparameters configured for it. The soil shear strength parameters from a standardized temporary sample set are used as input, and the corresponding safety factors from the standardized temporary sample set are used as output to train the temporary safety factor prediction surrogate model. SVR is a classic supervised learning algorithm in machine learning. Its core is to find an optimal hyperplane such that the deviation between the predicted and true values ​​for most samples does not exceed a predetermined value. In this embodiment, to balance evaluation efficiency and basic model performance, a set of fixed hyperparameter values ​​determined based on domain experience or preliminary experiments is used to configure the SVR model, rather than performing time-consuming hyperparameter optimization each time a temporary safety factor prediction surrogate model is built. For example, the penalty coefficient λ can be fixed at 1.0, and the parameter γ of the radial basis function kernel can be fixed at 0.1.

[0076] In a preferred embodiment, step S222, which uses cross-validation to calculate the prediction error of the temporary safety factor prediction proxy model, specifically includes the following steps:

[0077] The standardized temporary sample set is randomly divided into P mutually exclusive subsets;

[0078] Perform P rounds of verification sequentially; in each round of verification, perform the following steps:

[0079] Select one subset from the P subsets as the validation set for the current round, and combine the remaining P-1 subsets as the training set for the current round;

[0080] Using the preset hyperparameters, the support vector regression model is retrained with the current round training set to obtain the support vector regression model, and the support vector regression model is used as the validation model for the current round.

[0081] Taking P=5 as an example, in the first round, subset 1 is used as the validation set for the current round, and subsets 2-5 are combined as the training set for the current round. Model M1 is trained using fixed hyperparameters (e.g., penalty coefficient λ=1.0, radial basis function parameter γ=0.1). In the second round, subset 2 is used as the validation set for the current round, and subsets 1, 3, 4, and 5 are combined as the training set for the current round to train model M2. This process continues, resulting in a total of 5 models (M1 to M5). Each round trains a new model from scratch, strictly adhering to the cross-validation principle.

[0082] The current round's verification model is used to predict the current round's verification set to obtain the predicted safety coefficient;

[0083] The mean square error between the predicted safety coefficient and the inherent actual safety coefficient in the current round of the validation set is calculated as the validation error of the current round. For example, for the i-th round, the safety coefficients of 20 samples in the validation set are predicted using a support vector regression model to obtain 20 predicted safety coefficients. The mean square error between these 20 predicted safety coefficients and the corresponding 20 actual safety coefficients (which can be derived from the initial finite element simulation and are known data) is calculated.

[0084] Calculate the average of all P verification errors obtained in the P rounds of verification, and use the average as the prediction error of the temporary safety factor prediction surrogate model.

[0085] In a preferred embodiment, step S3 includes the following steps:

[0086] S31, Standardize the soil shear strength parameters in the training sample set to obtain a standardized training sample set;

[0087] S32, based on the standardized training sample set, a sparrow search algorithm is used to globally optimize the hyperparameters of the support vector regression model; wherein, the sparrow search algorithm is executed in each iteration: using the hyperparameter combination of the current iteration, a support vector regression model is constructed based on the standardized training sample set and cross-validated, and the calculated mean squared error is used as the fitness value of the hyperparameter combination of the current iteration; the sparrow search algorithm updates iteratively, searching for the hyperparameter combination that minimizes the fitness value as the optimal hyperparameter combination;

[0088] S33. Using the optimal hyperparameter combination and the standardized training sample set, the support vector regression model is trained to obtain the optimized safety coefficient prediction model.

[0089] This embodiment employs the Sparrow Search algorithm, using the standardized training sample set as a fixed data foundation, to automatically and globally optimize the hyperparameters (penalty coefficient λ and radial basis function kernel parameter γ) of the support vector regression model. The preset hyperparameter search space is set according to the empirical parameter range of the support vector regression model; for example, the search interval for the penalty coefficient λ is [0.1, 100], and the search interval for the radial basis function kernel parameter γ is [0.01, 1].

[0090] Specifically, this embodiment sets the population size of the sparrow search algorithm (e.g., 30 sparrows constitute a population), and the position of each sparrow is represented by a multi-dimensional vector. This vector corresponds to a set of hyperparameter combinations to be optimized and is initialized within a preset hyperparameter search space. In each iteration of the sparrow search algorithm, for each set of hyperparameter combinations in the population (i.e., the current position of each sparrow), the following operations are performed to calculate its fitness value:

[0091] Using this set of hyperparameters, a support vector regression model is trained based on the standardized training sample set. K-fold cross-validation is then performed on the support vector regression model, and the mean squared error (MSE) generated during the cross-validation process is calculated. This MSE is used as the fitness value of the hyperparameter combination in the current iteration. The calculation method is the same as the cross-validation method described earlier and will not be repeated here. A higher fitness value indicates that the optimization goal of the sparrow search algorithm is to find the hyperparameter combination that minimizes this fitness value during the iteration process. This embodiment achieves deep mining of limited sample data, replacing the traditional, experience-dependent, inefficient grid search or manual trial and error.

[0092] In a preferred embodiment, step S4 includes the following steps:

[0093] S41, within the effective value range, the parameters are discretized at equal intervals along the dimensions of cohesion and internal friction angle, respectively, to generate a sequence of cohesion parameter values ​​and a sequence of internal friction angle parameter values; the effective value range is a physically reasonable interval determined based on engineering geological survey data. For example, for a certain reservoir bank landslide, the cohesion c∈[10,50] is set, with units of kPa, and the internal friction angle... ∈[15,30], unit: °; The discretization process can use a linear interpolation method to uniformly generate multiple values ​​within the interval of cohesion c, forming a sequence of cohesion parameter values. In the internal friction angle Multiple values ​​are generated within the interval, forming a sequence of internal friction angle parameter values.

[0094] S42, the cohesion parameter value sequence and the internal friction angle parameter value sequence are combined pairwise to generate a two-dimensional grid point set covering the effective value range; the cohesion parameter value sequence and the internal friction angle parameter value sequence are combined pairwise (i.e., Cartesian product) to generate a two-dimensional Cartesian grid covering the entire effective value range. Each grid point is represented by coordinates (cohesion, internal friction angle), representing a set of potential soil shear strength parameters to be evaluated.

[0095] S43, input the cohesion and internal friction angle corresponding to each grid point in the two-dimensional grid point set into the safety factor prediction model; calculate the input cohesion and internal friction angle through the safety factor prediction model, output the predicted safety factor corresponding to each grid point, and obtain a set of predicted safety factors corresponding to each grid point; input the parameter values ​​(cohesion, internal friction angle) of each grid point in the two-dimensional grid point set obtained in step S42 into the already trained safety factor prediction model in batches to quickly obtain the predicted safety factor corresponding to each grid point.

[0096] S44, calculate the absolute error between each predicted safety factor in the predicted safety factor set and the target safety factor; find the grid point with the smallest absolute error from the two-dimensional grid point set, and use the cohesion and internal friction angle corresponding to this grid point as the soil shear strength parameter inversion result matching the target safety factor. Specifically, by traversing the predicted safety factor set, for each predicted safety factor, calculate its absolute error with the target safety factor. This can be done by calling the minimum value index function to search for the grid point with the smallest absolute error in the two-dimensional grid point set, extracting the corresponding coordinates (cohesion, internal friction angle) from the grid point with the smallest absolute error, and outputting them as the final soil shear strength parameter inversion result.

[0097] like Figure 2 The scatter plot shown below illustrates the spatial distribution of soil shear strength parameters. The horizontal axis represents cohesion (c), and the vertical axis represents the internal friction angle. In this diagram, the red dots represent the final soil shear strength parameter inversion results, and the blue dots represent other grid points in the two-dimensional grid point set.

[0098] In addition, to further illustrate the technical solution of this application, a comparison was made between the inversion results of conventional orthogonal design and uniform design and that of this application:

[0099] The inversion results of orthogonal design, uniform design, and this application show significant differences in uniformity, efficiency, accuracy, and nonlinear adaptability. The core conclusions are: this application has outstanding adaptability in complex regions, the surrogate model has the best inversion accuracy, the uniform design balances efficiency and uniformity, and the orthogonal design is suitable for linear scenarios.

[0100] This application performs cross-validation on the training samples during training and calculates the average mean squared error of each round of cross-validation. Using the mean squared error as the inversion objective function, the inversion results are shown in Table 1:

[0101] Table 1: Comparison of inversion results between orthogonal design, uniform design and this application

[0102]

[0103] To verify the accuracy of the inversion results, the inversion results were substituted into the slope stability analysis software Slide to obtain the safety factor, and the relative error between the safety factor and the original data was calculated. The results are shown in Table 2. It can be seen that the relative error based on the method of this application is only 0.027%, which is much higher than that of orthogonal design (0.218%) and uniform design (0.403%).

[0104] Table 2: Safety factors of orthogonal design, uniform design and this application, and relative errors with the original data.

[0105]

[0106] Furthermore, the mean relative error varies with different sample sizes as follows:

[0107] By gradually increasing the sample size, the average relative error of the inversion results was calculated for each corresponding sample size. The relationship between the sample size and the average relative error is as follows: Figure 4 As shown.

[0108] from Figure 4 It can be seen that the mean relative error decreases significantly with the increase of the sample size. When the sample size is small (e.g., 15), the mean relative error exceeds 6%; as the sample size increases to over 55, the mean relative error rapidly decreases to below 1%; when the sample size reaches 95 or more, the mean relative error tends to stabilize and approach 0%. This indicates that increasing the sample size can effectively reduce the mean relative error of the inversion, and when the sample size is large enough (e.g., over 95), the error can be controlled at an extremely low level and tends to stabilize.

[0109] The present invention also provides a device for inverting the shear strength parameters of landslide soil on reservoir banks, used to perform the inversion method for the shear strength parameters of landslide soil on reservoir banks as described above, including:

[0110] The target input module is used to obtain the target safety factor.

[0111] The sample generation module is used to determine the effective range of soil shear strength parameters and generate a training sample set through numerical simulation within the effective range; wherein, the training sample set includes multiple sets of soil shear strength parameters and corresponding safety factors.

[0112] The model building module is used to build a safety factor prediction model based on the training sample set using machine learning methods; wherein, the safety factor prediction model includes the mapping relationship between soil shear strength parameters and safety factors;

[0113] The inversion search module is used to search within the effective value range based on the safety factor prediction model and the target safety factor to determine the soil shear strength parameter inversion result that matches the target safety factor.

[0114] The present invention also provides a device for inverting the shear strength parameters of soil in a reservoir bank landslide, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described method for inverting the shear strength parameters of soil in a reservoir bank landslide.

[0115] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for inverting the shear strength parameters of landslide soil on the reservoir bank.

[0116] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0117] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A bank landslide soil body shear strength parameter inversion method, characterized in that, The method comprises the following steps: S1, obtaining a target safety factor; S2, determining an effective value range of soil shear strength parameters, and generating a training sample set through numerical simulation within the effective value range; wherein the training sample set comprises multiple groups of soil shear strength parameters and corresponding safety factors; S3, constructing a safety factor prediction model through a machine learning method according to the training sample set; wherein the safety factor prediction model comprises a mapping relationship between soil shear strength parameters and safety factors; S4, searching in the effective value range according to the safety factor prediction model and the target safety factor, and determining a soil shear strength parameter inversion result matched with the target safety factor; The step S3 comprises the following steps: S31, performing standardization processing on the soil shear strength parameters in the training sample set to obtain a standardized training sample set; S32, based on the standardized training sample set, using a sparrow search algorithm to globally optimize hyperparameters of a support vector regression model; wherein the sparrow search algorithm performs the following in each iteration: using the hyperparameter combination of the current iteration, constructing a support vector regression model based on the standardized training sample set and performing cross-validation to calculate the mean square error as the fitness value of the hyperparameter combination of the current iteration; the sparrow search algorithm searches for the hyperparameter combination that minimizes the fitness value as the optimal hyperparameter combination through iterative updating; S33, using the optimal hyperparameter combination and the standardized training sample set to train the support vector regression model to obtain an optimized safety factor prediction model; The step S4 comprises the following steps: S41, discretizing the effective value range at equal intervals in the dimensions of cohesion and internal friction angle respectively to generate a cohesion parameter value sequence and an internal friction angle parameter value sequence; S42, combining the cohesion parameter value sequence and the internal friction angle parameter value sequence two by two to generate a two-dimensional grid point set covering the effective value range; S43, inputting the cohesion and internal friction angle corresponding to each grid point in the two-dimensional grid point set into the safety factor prediction model; calculating the input cohesion and internal friction angle through the safety factor prediction model to output the predicted safety factor corresponding to each grid point, thereby obtaining a predicted safety factor set corresponding to the grid points one by one; S44, calculating the absolute error between each predicted safety factor in the predicted safety factor set and the target safety factor; finding the grid point with the smallest absolute error from the two-dimensional grid point set, and taking the cohesion and internal friction angle corresponding to the grid point as the soil shear strength parameter inversion result matched with the target safety factor.

2. The method according to claim 1, wherein, The step S2 comprises the following steps: S21, uniformly generating N groups of initial soil shear strength parameter sample points in the effective value range, performing finite element numerical simulation of slope stability for each group of initial soil shear strength parameter sample points, outputting corresponding initial safety factors, and collecting all initial soil shear strength parameter sample points and corresponding initial safety factors to obtain an initial training sample set; S22, taking the initial training sample set as a current training sample set, and performing K rounds of iterative sampling; each round of iteration includes steps S221-S223: S221, constructing a Thiessen polygon in the effective value range according to all soil shear strength parameter sample points in the current training sample set, and dividing the effective value range into multiple Thiessen polygon units; each Thiessen polygon unit contains one soil shear strength parameter sample point; S222, randomly generating M candidate soil shear strength parameter sample points in the effective value range; for each candidate soil shear strength parameter sample point, adding it to the current training sample set to form a temporary sample set, constructing a temporary safety factor prediction proxy model according to the temporary sample set, calculating the prediction error of the temporary safety factor prediction proxy model by using a cross-validation method, and recording the prediction error as the evaluation error of the candidate soil shear strength parameter sample point; S223, selecting a candidate soil shear strength parameter sample point with the largest evaluation error from the M candidate soil shear strength parameter sample points as a current round optimal supplementary sample point, performing finite element numerical simulation of slope stability on the current round optimal supplementary sample point, outputting a corresponding supplementary safety factor, and adding the current round optimal supplementary sample point and the corresponding supplementary safety factor to the current training sample set to update the current training sample set; S23, after completing K rounds of iterative sampling, taking the finally updated current training sample set as the training sample set.

3. The method according to claim 2, wherein, The step S222 of constructing a temporary safety factor prediction proxy model according to the temporary sample set includes the following steps: standardizing the soil shear strength parameters in the temporary sample set to obtain a standardized temporary sample set; using a support vector regression algorithm, configuring a set of preset hyperparameters for the support vector regression algorithm, taking the standardized soil shear strength parameters in the standardized temporary sample set as input, and taking the corresponding safety factors in the standardized temporary sample set as output, and training to obtain the temporary safety factor prediction proxy model.

4. The method according to claim 3, wherein, The step S222 of calculating the prediction error of the temporary safety factor prediction proxy model by using a cross-validation method includes the following steps: randomly dividing the standardized temporary sample set into P mutually exclusive subsets; performing P rounds of validation in turn; in each round of validation, the following steps are performed: selecting one subset from the P subsets as a current round validation set, and combining the remaining P-1 subsets as a current round training set; using the preset hyperparameters to retrain a support vector regression model using the current round training set, and taking the support vector regression model as a current round validation model; A prediction safety factor is obtained by using a current round verification model to predict the current round verification set; A mean square error between the prediction safety factor and an inherent actual safety factor in the current round verification set is calculated as a verification error of the current round; An average of all P verification errors obtained in P round verifications is calculated, and the average is taken as a prediction error of the temporary safety factor prediction agent model.

5. The method according to claim 2, wherein, The number N of initial soil shear strength parameter sample points ranges from 5 to 20, the number K of iteration sampling rounds ranges from 50 to 200, and M ranges from 800 to 1200.

6. A device for back-calculation of shear strength parameters of reservoir bank landslide soil mass, for performing the method for back-calculation of shear strength parameters of reservoir bank landslide soil mass according to any one of claims 1-5, characterized in that, Comprise: A target input module for obtaining a target safety factor; A sample generation module for determining an effective value range of soil shear strength parameters and generating a training sample set in the effective value range through numerical simulation; wherein the training sample set comprises multiple groups of soil shear strength parameters and corresponding safety factors; A model construction module for constructing a safety factor prediction model through a machine learning method according to the training sample set; wherein the safety factor prediction model comprises a mapping relationship between soil shear strength parameters and safety factors; An inversion search module for searching in the effective value range according to the safety factor prediction model and the target safety factor to determine an inversion result of soil shear strength parameters matched with the target safety factor.

7. A device for back-calculation of soil shear strength parameters of a bank landslide, characterized in that, A computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the bank landslide soil shear strength parameter inversion method according to any one of claims 1 to 5.

8. A storage medium storing a computer program, characterized by The computer program is executed by the processor to implement the steps of the bank landslide soil shear strength parameter inversion method according to any one of claims 1 to 5.

Images