Slope stability probability prediction method and system and slope stability probability regression model construction system

Abstract

The invention discloses a slope stability probability prediction method and system and a slope stability probability regression model construction system. Comprising the following steps: firstly, collecting multiple parts of slope monitoring data as samples to construct a data set; and then, a probability regression model is constructed in combination with a LightGBM model and a GAMLSS model, the probability regression model is trained through a data set, and an optimization algorithm is fused to carry out hyper-parameter optimization on the probability regression model, so that optimal probability regression model parameters are obtained. The constructed probability regression model is combined with the LightGBM model and the GAMLSS model to construct the probability regression model, on the premise that the slope stability prediction precision is ensured, any uncertainty contained in the slope stability is accurately quantified, and the slope stability evaluation accuracy is improved. Parameters of the probability regression model are optimized by adopting an optimization algorithm, the prediction precision and reliability of the probability regression model are further improved, the slope stability evaluation accuracy is improved, and rich decision information is provided for slope prevention and control design and risk management.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing systems or methods specially suitable for prediction purposes, and in particular to a slope stability probability prediction method and system and a slope stability probability regression model construction system. Background Art

[0002] Mining, water conservancy, construction, transportation and other departments are all involved in a large number of slope stability issues. Slope stability judgment is the basis of engineering risk management and prevention and control design. Therefore, in-depth research on slope stability evaluation and prediction methods is particularly important. At present, slope stability evaluation methods include engineering geological analogy method, limit equilibrium method, numerical method and evaluation methods based on statistical theory or machine learning.

[0003] Among them, methods based on statistical principles or machine learning are widely favored in slope stability evaluation and landslide prediction because they have strong ability to deal with complex nonlinear problems and fully consider the impact of different factors on slope stability. For example, gray system theory, fuzzy hierarchical discriminant method, BP neural network, support vector machine SVM, decision tree, Adaboost algorithm, XGBoost algorithm, random forest RF algorithm and deep learning.

[0004] The slope system is a highly complex nonlinear system, which is affected by many internal and external factors such as rainfall, earthquakes, human activities, stratum lithology, geological structure, hydrological conditions, etc., which leads to strong randomness and uncertainty in slope stability. Although machine learning methods have achieved good results in slope stability evaluation and prediction, they cannot accurately reflect and quantify the inherent randomness in slope stability evaluation, resulting in incorrect estimation of the true stability state of the slope. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention proposes a slope stability probability prediction method, system and slope stability probability regression model construction system, which can accurately quantify the arbitrary uncertainty contained in the slope stability while ensuring the accuracy of slope stability prediction. The specific technical solution is as follows: In a first aspect, a method for constructing a slope stability probabilistic regression model is provided. In a first implementable manner of the first aspect, the method comprises: Obtain multiple slope monitoring data to build a data set; The probability regression model constructed by combining the LightGBM model and the GAMLSS model is trained through the data set, and the model parameters of the probability regression model are optimized by using an optimization algorithm.

[0006] In combination with the first implementable manner of the first aspect, in a second implementable manner of the first aspect, acquiring multiple slope monitoring data to construct a data set includes: The acquired slope monitoring data is normalized.

[0007] In combination with the first implementable manner of the first aspect, in a third implementable manner of the first aspect, the log-likelihood / loss function used in the constructed probability regression model is: ; in, is the number of iterations, and are the first and second derivatives of the loss function, respectively. For the LightGBM model, is the leaf node weight, is the penalty item, is the total number of leaf nodes, is the target number, is the regularization parameter.

[0008] In combination with the first implementable manner of the first aspect, in a fourth implementable manner of the first aspect, the constructed probability regression model adopts a mixed distribution to describe the slope stability coefficient.

[0009] In combination with the first implementable manner of the first aspect, in a fifth implementable manner of the first aspect, a PSO algorithm is used to perform hyperparameter optimization on the probabilistic regression model.

[0010] In a second aspect, a slope stability probability prediction method is provided, which is characterized by comprising: A probabilistic regression model is constructed by using the slope stability probabilistic regression model construction method as described in any one of the first to fifth possible implementations of the first aspect; Slope monitoring data of the slope to be predicted is obtained, and the slope stability probability of the slope to be predicted is predicted through a probability regression model.

[0011] In conjunction with the first implementable manner of the second aspect, in a second implementable manner of the second aspect, predicting the slope stability probability of the slope to be predicted includes: The slope monitoring data of the slope to be predicted is normalized.

[0012] In combination with the first implementable manner of the second aspect, in a third implementable manner of the second aspect, the method further includes: The SHAP method was used to perform interpretability analysis on the slope stability probability in order to quantify the contribution of different factors to slope stability.

[0013] In a third aspect, a slope stability probability regression model construction system is provided, comprising: A data set construction module, configured to acquire multiple slope monitoring data to construct a data set; The prediction model construction module is configured to train the probability regression model constructed by combining the LightGBM model and the GAMLSS model through the data set, and optimize the model parameters of the probability regression model using an optimization algorithm.

[0014] In a fourth aspect, a slope stability probability prediction system is provided, comprising: A model building module, configured to construct a probabilistic regression model using the slope stability probabilistic regression model building method as described in any one of the first to fifth possible implementations of the first aspect; The probability prediction module is configured to obtain slope monitoring data of the slope to be predicted, and predict the slope stability probability of the slope to be predicted through a probability regression model.

[0015] Beneficial effects: The slope stability probability prediction method, system and slope stability probability regression model construction system of the present invention can accurately quantify the arbitrary uncertainty contained in the slope stability while ensuring the accuracy of slope stability prediction by integrating the LightGBM model and the GAMLSS model. By optimizing the parameters of the probability regression model using an optimization algorithm, the prediction accuracy and reliability of the probability regression model can be further improved, thereby building a reliable slope stability prediction model and improving the accuracy of slope stability evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific implementation of the present invention, the following will briefly introduce the drawings required for use in the specific implementation. In all the drawings, each element or part is not necessarily drawn according to the actual scale.

[0017] Figure 1 A flow chart of a method for constructing a slope stability probabilistic regression model provided by an embodiment of the present invention; Figure 2 A flow chart of a slope stability probability prediction method provided by an embodiment of the present invention; Figure 3 A system block diagram of a slope stability probabilistic regression model construction system provided by an embodiment of the present invention; Figure 4 A system block diagram of a slope stability probability prediction system provided by an embodiment of the present invention; Figure 5 An optimal estimation diagram of a mixed distribution of safety factors of a slope provided by an embodiment of the present invention; Figure 6This is a schematic diagram of the results of stability prediction using the prediction method provided by the present invention; Figure 7 for Figure 6 Probability prediction results for cases 74, 76, 79 and 81; Figure 8 It is the importance ranking diagram of the factors affecting slope stability; Fig. 9 A SHAP dependency graph of the slope angle calculated based on the prediction result obtained by the prediction method provided by the present invention; Fig.10 The SHAP dependency graph of the slope height obtained by calculating the prediction results based on the prediction method provided by the present invention. DETAILED DESCRIPTION

[0018] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0019] like Figure 1 The flowchart of the slope stability probability regression model construction method shown in FIG. 1 includes: Step 1, obtain multiple slope monitoring data to build a data set; Step 2: Train the probability regression model constructed by combining the LightGBM model and the GAMLSS model through the data set, and optimize the model parameters of the probability regression model using an optimization algorithm.

[0020] Specifically, first, we can collect multiple different slope monitoring data as samples to build a data set. Then, we can combine the LightGBM model and the GAMLSS model to build a probabilistic regression model, determine the target variable, that is, the optimal distribution of the slope stability coefficient, and train the probabilistic regression model through the constructed data set to obtain a trained probabilistic regression model. During the training process, we can integrate the optimization algorithm to optimize the hyperparameters of the probabilistic regression model and obtain the optimal probabilistic regression model parameters.

[0021] The probability regression model constructed by combining the LightGBM model and the GAMLSS model can accurately quantify the arbitrary uncertainty contained in the slope stability and improve the accuracy of the slope stability evaluation under the premise of ensuring the accuracy of slope stability prediction. And by using the optimization algorithm to optimize the parameters of the probability regression model, the prediction accuracy and reliability of the probability regression model can be further improved, and the accuracy of slope stability evaluation can be improved, so as to provide rich decision-making information for slope prevention and control design and risk management.

[0022] In this embodiment, optionally, multiple sets of slope monitoring data are obtained to construct a data set, including: The acquired slope monitoring data is normalized.

[0023] Specifically, the collected slope monitoring data include factors affecting slope stability, such as bulk density, cohesion, internal friction angle, slope angle, slope height, pore pressure ratio, etc. The data values ​​corresponding to each factor in the acquired slope monitoring data can be normalized, and the specific calculation formula is as follows: ; in, , are the values ​​before and after factor normalization, , They are the maximum and minimum values ​​corresponding to the factors in the slope monitoring data, respectively.

[0024] Specifically, the probabilistic regression model combines the LightGBM model and the GAMLSS model. It models the distribution of the target variable to solve the uncertainty problem contained in objective things and achieve the purpose of probabilistic regression or probabilistic prediction. When building a probabilistic regression model, it is necessary to select a suitable log-likelihood / loss function and determine the optimal distribution of the target variable.

[0025] In this embodiment, optionally, the log-likelihood / loss function used in the constructed probability regression model is: ; in, is the number of iterations, and are the first and second derivatives of the loss function, respectively. For the LightGBM model, is the leaf node weight, is the penalty item, is the total number of leaf nodes, is the target number, is the regularization parameter.

[0026] The GAMLSS model is to model the distribution of the target variable. The specific expression is as follows: ; in, The target variable The distribution of is a distribution parameter, and each distribution parameter is modeled by an additive model, such as a linear model, spline regression, etc. In this embodiment, the LightGBM model is used to replace the additive model in the GAMLSS model, and the gradient boosting tree of the LightGBM model is used to fit each parameter, that is, the known function between the distribution parameter and the predictor. is a known function between the distribution parameter and the prediction factor. The specific expression is as follows: .

[0027] in, is the predictor, i.e. the LightGBM model.

[0028] It can be seen that it is extremely important to determine the appropriate distribution type of the target variable.

[0029] In this embodiment, optionally, the constructed probability regression model adopts a mixed distribution to describe the slope stability coefficient.

[0030] Specifically, the mixed distribution can be used to describe the slope stability coefficient, and the specific expression is as follows: ; in, is the mixed distribution of observations, is the mth sub-distribution, each sub-distribution has its own parameters and weight , The mixed distribution is composed of multiple different sub-distributions, and the sub-distribution can be a simple distribution type such as Normal distribution or StudentT distribution. Based on the stability coefficient in the training set, the mixed distribution and parameters of the stability coefficient of the present invention are finally determined as follows: Figure 5 shown.

[0031] In this way, the constructed probabilistic regression model can accurately quantify the inherent uncertainty contained in slope stability, thereby providing rich decision-making information for slope prevention design and risk management. In this embodiment, optionally, a PSO algorithm is used to perform hyperparameter optimization on the probabilistic regression model.

[0032] Specifically, the particle swarm algorithm can be used to optimize the parameters of the probability regression model. Specifically, first, the parameters of the particle swarm algorithm can be initialized according to the constructed probability regression model, including: eta, max_depth, min_gain_to_split, min_sum_hessian_in_leaf, subsample, feature_fraction and boosting. The initialization parameter value ranges are [1e-5, 1], [1, 200], [1e-8, 20], [1e-8, 500], [0.2, 1], [0.2, 1] and [categorical, gbdt]. Then, determine the population size, position range and speed range, as well as the position of the initialized particles. and speed , set the maximum number of iterations and initialize the learning factor and After that, iterative training is carried out to compare the fitness values ​​to update the optimal solution vector of a single particle and the global optimal solution vector of the particle swarm, so as to obtain the optimal probability regression model parameters. The specific calculation formula is as follows: ; ; ; in, is the fitness function, is the number of training samples, is the particle velocity, is the particle position, , is a random number between 0 and 1. is the true value of the sample, is the predicted value of the probabilistic regression model, is the parameter to be optimized, is the individual extreme value.

[0033] It should be understood that this embodiment is only illustrated by the particle swarm algorithm, but the present invention is not limited thereto, and other optimization algorithms may also be used to optimize the parameters of the probabilistic regression model, such as the Bayesian optimization algorithm, the genetic optimization algorithm, and the like.

[0034] In order to verify the prediction effect of the probability regression model constructed by the present invention, the stability coefficients of multiple slopes were predicted using the constructed probability regression model. The prediction results are as follows: Figure 6 , Figure 7As shown, the predicted probability density distributions of cases 74, 76, 79 and 81 are all close to Gaussian distributions, and the true stability coefficient of the slope is within the 0.05 and 0.95 quantile intervals of the predicted distribution, and is around the uniform distribution and predicted mean. This shows that the probability regression model constructed using the construction method of the present invention can obtain high-quality probability prediction results, accurately quantify the inherent uncertainty and randomness in slope stability, and provide rich decision-making information for slope prevention design and risk management.

[0035] like Figure 2 The flowchart of the slope stability probability prediction method shown in FIG. 1 includes: Step S1, using the above slope stability probability regression model construction method to construct a probability regression model; Step S2: Obtain slope monitoring data of the slope to be predicted, and predict the slope stability probability of the slope to be predicted by using a probabilistic regression model.

[0036] Specifically, first, the above-mentioned construction method can be used to construct a probabilistic regression model for predicting the slope stability coefficient. Then, the slope monitoring data of the slope to be predicted can be obtained, and the stability probability of the slope can be predicted by the constructed probabilistic regression model.

[0037] In this embodiment, optionally, predicting the slope stability probability of the slope to be predicted includes: normalizing the slope monitoring data of the slope to be predicted.

[0038] In this embodiment, optionally, it further includes: The SHAP method was used to perform interpretability analysis on the slope stability probability in order to quantify the contribution of different factors to slope stability.

[0039] Specifically, after obtaining the stability probability of the slope to be predicted through the probabilistic regression model, the SHAP method can be used to perform an interpretable analysis of the stability probability. The calculated SHAP value quantifies the contribution of different factors to slope stability, thereby providing more abundant decision-making information for slope safety construction and risk management. The specific calculation formula of the SHAP value is as follows: ; in, is the initial expected value, is the number of input features, is the SHAP value of the feature, Indicates whether the corresponding feature can be observed. The specific calculation formula of the SHAP value of the feature is as follows: ; in, is the set of all features in the training set, for A subset of , Based on S and The calculated sample mean is To include and does not include The difference of .

[0040] In this embodiment, the SHAP method can be used to calculate the contribution of slope angle, slope height, bulk density, cohesion, internal friction angle and pore pressure ratio to the slope stability coefficient, and the results are shown in the order of feature importance. Figure 8 As shown, it can be seen that the slope angle and slope height have the greatest influence on the slope stability, followed by cohesion, while the bulk density, internal friction angle and pore pressure ratio have the worst influence on the slope stability, especially the pore pressure ratio, whose contribution to the model output is 0, which is in line with the actual physical laws followed by the slope, proving the accuracy and effectiveness of the present invention.

[0041] In order to analyze the detailed contribution of different input features to the model prediction results, the SHAP method is used to calculate the single factor dependency graph, such as Fig. 9 , Fig.10 It can be seen that with the increase of slope angle and slope height, the SHAP value gradually decreases, indicating that the greater the slope gradient and height, the lower the stability of the slope, which is consistent with the actual physical law followed by the slope, further proving the accuracy and effectiveness of the present invention.

[0042] like Figure 3 The system block diagram of the slope stability probability regression model construction system shown in FIG. 1 includes: A data set construction module, configured to acquire multiple slope monitoring data to construct a data set; The prediction model construction module is configured to train the probability regression model constructed by combining the LightGBM model and the GAMLSS model through the data set, and optimize the model parameters of the probability regression model using an optimization algorithm.

[0043] Specifically, the construction system includes a data set construction module and a prediction model construction module, wherein the data set construction module can collect multiple different slope monitoring data as samples to construct a data set. The prediction model construction module can combine the LightGBM model and the GAMLSS model to construct a probability regression model, and determine the target variable, that is, the optimal distribution of the slope stability coefficient, and train the probability regression model through the constructed data set to obtain a trained probability regression model. During the training process, the optimization algorithm can be integrated to optimize the hyperparameters of the probability regression model to obtain the optimal probability regression model parameters.

[0044] The probability regression model constructed by combining the LightGBM model and the GAMLSS model can accurately quantify the arbitrary uncertainty contained in the slope stability and improve the accuracy of the slope stability evaluation under the premise of ensuring the accuracy of slope stability prediction. And by using the optimization algorithm to optimize the parameters of the probability regression model, the prediction accuracy and reliability of the probability regression model can be further improved, and the accuracy of slope stability evaluation can be improved, so as to provide rich decision-making information for slope prevention and control design and risk management.

[0045] like Figure 4 The system block diagram of the slope stability probability prediction system shown in FIG. 1 includes: A model building module is configured to use the above slope stability probability regression model building method to build a probability regression model; The probability prediction module is configured to obtain slope monitoring data of the slope to be predicted, and predict the slope stability probability of the slope to be predicted through a probability regression model.

[0046] Specifically, the prediction system includes a model building module and a probability prediction module. The model building module can use the above-mentioned construction method to build a probability regression model for predicting the slope stability coefficient. The probability prediction module can obtain the slope monitoring data of the slope to be predicted, and predict the stability probability of the slope through the constructed probability regression model.

[0047] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A slope stability probability prediction method, characterized in that: include: Obtain multiple slope monitoring data to build a data set; The probability regression model constructed by combining the LightGBM model and the GAMLSS model is trained by using the data set, and the model parameters of the probability regression model are optimized by using an optimization algorithm; Slope monitoring data of the slope to be predicted is obtained, and the slope stability probability of the slope to be predicted is predicted through a probability regression model.

2. The slope stability probability prediction method according to claim 1 is characterized in that: Obtain multiple slope monitoring data to build a data set, including: The acquired slope monitoring data is normalized.

3. The slope stability probability prediction method according to claim 1 is characterized in that: The log-likelihood / loss function used in the constructed probability regression model is: ； in, is the number of iterations, and are the first and second derivatives of the loss function, respectively. For the LightGBM model, is the leaf node weight, is the penalty item, is the total number of leaf nodes, is the target number, is the regularization parameter.

4. The slope stability probability prediction method according to claim 1 is characterized in that: The constructed probabilistic regression model adopts a mixed distribution to describe the slope stability coefficient.

5. The slope stability probability prediction method according to claim 1 is characterized in that: The PSO algorithm is used to optimize the hyperparameters of the probabilistic regression model.

6. The slope stability probability prediction method according to claim 1 is characterized in that: Predicting the slope stability probability of the slope to be predicted, including: The slope monitoring data of the slope to be predicted is normalized.

7. The slope stability probability prediction method according to claim 1 is characterized in that: Also includes: The SHAP method was used to perform interpretability analysis on the slope stability probability in order to quantify the contribution of different factors to slope stability.

8. A slope stability probability regression model construction system, characterized in that: include: A data set construction module, configured to acquire multiple slope monitoring data to construct a data set; The prediction model construction module is configured to train the probability regression model constructed by combining the LightGBM model and the GAMLSS model through the data set, and optimize the model parameters of the probability regression model using an optimization algorithm.

9. A slope stability probability prediction system, characterized in that: include: Model building module, configured to obtain multiple slope monitoring data to build a data set; The probability regression model constructed by combining the LightGBM model and the GAMLSS model is trained by using the data set, and the model parameters of the probability regression model are optimized by using an optimization algorithm; The probability prediction module is configured to obtain slope monitoring data of the slope to be predicted, and predict the slope stability probability of the slope to be predicted through a probability regression model.

Images