Slope instability early warning method and system

Abstract

The invention relates to the technical field of data processing, and discloses a slope instability early warning method and system. The method comprises the steps of performing normalization processing on slope monitoring data through polar coordinate deviation standardization, performing expansion processing on minority class samples according to synthetic minority class oversampling, performing optimization processing on extreme gradient lifting model parameters through Newton-Raphson optimization, and performing performance evaluation through ten-fold cross validation. And performing weight analysis processing on the features according to interpretability analysis to obtain slope instability risk early warning output. According to the method, the problems of single data preprocessing method, unbalanced samples, insufficient model optimization and lack of interpretability in the existing slope instability early warning technology are solved. According to the method, the accuracy, the stability and the interpretability of slope instability early warning are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a slope instability early warning method and system. Background Art

[0002] Slope instability early warning is an important technology in the field of geotechnical engineering and geological disaster prevention and control. Existing technologies mainly use a single sensor monitoring method to obtain slope deformation data by installing displacement meters, inclinometers, strain gauges and other equipment, and perform instability early warning based on threshold judgment or simple statistical analysis. Some technologies also combine traditional machine learning algorithms such as support vector machines and neural networks to analyze and process monitoring data, and realize the early warning function by establishing a mapping relationship between input features and instability status.

[0003] Existing slope instability early warning technology has significant shortcomings. First, the data preprocessing capability is limited, and it is unable to effectively process multi-source monitoring data of different dimensions and numerical ranges, resulting in the loss of important feature information and model training deviation. Secondly, the sample imbalance problem is serious. Since slope instability events are relatively rare, the number of unstable samples is far less than that of stable samples. Traditional methods find it difficult to fully learn the characteristics of instability patterns, and the prediction accuracy is not high. In addition, the existing algorithms lack an effective hyperparameter optimization mechanism, and the model performance is difficult to reach the optimal state. At the same time, there is a lack of model interpretability analysis, and the key factors affecting slope stability cannot be identified. The early warning results lack a scientific basis.

[0004] Based on the above analysis, existing technologies suffer from a single method for data standardization. Polar coordinate deviation normalization, a new data preprocessing method, has not yet been applied to slope instability early warning. Effectively converting multidimensional geological parameters into a unified feature representation remains a technical challenge. Regarding sample balancing, although synthetic minority oversampling has been applied in other fields, the generation of representative synthetic instability samples, given the particularity and complexity of slope instability samples, requires further research. Regarding model optimization, the combination of the Newton-Raphson optimization algorithm and the extreme gradient boosting model has not been fully explored. Improving the performance of slope stability discrimination models using second-order optimization information is an urgent technical issue. Regarding model validation, although ten-fold cross-validation is a mature evaluation method, designing a reasonable validation strategy to ensure model generalization in the specific context of slope instability early warning remains a challenge. Regarding interpretability analysis, existing technologies lack effective methods for quantifying feature importance, and the application of Shapley value calculation in slope engineering remains unresolved. Accurately identifying key influencing factors of slope instability based on game theory principles and forming a scientific early warning decision-making mechanism remains a bottleneck in current technological development. Summary of the Invention

[0005] This application provides a slope instability early warning method and system that addresses the problems of existing slope instability early warning technologies, such as single data preprocessing methods, sample imbalance, insufficient model optimization, and lack of interpretability. This application improves the accuracy, stability, and interpretability of slope instability early warnings.

[0006] In the first aspect, the present application provides a slope instability warning method, which includes: normalizing the original slope monitoring data through polar coordinate deviation standardization to obtain a standardized slope feature vector; expanding the minority class samples in the standardized slope feature vector according to synthetic minority class oversampling to obtain a slope training data set; iteratively optimizing the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model by Newton-Raphson optimization to obtain a slope stability discrimination model; calculating the accuracy and precision of the slope training data set and the slope stability discrimination model through ten-fold cross validation to obtain a slope stability discrimination model; and performing weight parsing on the gravity density, cohesion, and internal friction angle features in the slope stability discrimination model according to interpretability analysis to obtain a slope instability risk warning output.

[0007] In a second aspect, the present application provides a slope instability early warning system, the slope instability early warning system comprising: A processing module is used to normalize the original data of slope monitoring by polar coordinate deviation normalization to obtain a standardized slope characteristic vector; An expansion module, configured to expand minority class samples in the standardized slope feature vector according to synthetic minority class oversampling to obtain a slope training data set; The optimization module is used to iteratively optimize the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model using Newton-Raphson optimization to obtain a slope stability discrimination model; a calculation module, configured to calculate the accuracy and precision of the slope training data set and the slope stability discrimination model through ten-fold cross validation to obtain the slope stability discrimination model; The analytical module is used to perform weighted analytical processing on the gravity density, cohesion, and internal friction angle characteristics in the slope stability discrimination model based on interpretability analysis to obtain a slope instability risk warning output.

[0008] In a third aspect, a slope instability warning device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the slope instability warning device executes the above-mentioned slope instability warning method.

[0009] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on the computer, the computer executes the above-mentioned slope instability early warning method.

[0010] In the technical solution provided by the present application, the original data of slope monitoring is normalized by polar coordinate deviation normalization technology, which effectively solves the problem of unified representation of parameters of different dimensions such as gravity density, cohesion, internal friction angle, slope height, slope gradient, etc. Compared with the traditional linear normalization method, polar coordinate transformation can maintain the spatial relationship characteristics between parameters and avoid the loss of data information. At the same time, the minority class samples in the standardized slope feature vector are expanded by synthetic minority class oversampling, which fundamentally solves the class imbalance problem caused by the scarcity of slope instability samples. The generated synthetic instability samples not only increase the diversity of training data, but also maintain the characteristic distribution of the original instability mode, significantly improving the model's learning ability for instability samples. The Newton-Raphson optimization algorithm has an important influence on the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model. The method performs iterative optimization on the slope stability discriminant model and achieves faster convergence speed and higher parameter optimization accuracy than the traditional first-order optimization method by utilizing the second-order gradient information, so that the slope stability discrimination model can reach the optimal performance state. The accuracy and precision of the slope training data set and the slope stability discrimination model are calculated and processed by ten-fold cross-validation. The reliability and generalization ability of the model performance evaluation are ensured through multiple random divisions and repeated verifications, and the accidental errors that may be caused by single training and verification are avoided. The interpretability analysis performs weighted parsing on the gravity density, cohesion, and internal friction angle characteristics in the slope stability discrimination model. The key factors affecting slope stability are accurately identified based on the Shapley value calculation, providing a scientific basis for early warning decision-making. The overall technical solution has achieved a comprehensive improvement in the accuracy, stability and interpretability of slope instability warning.

[0011] The angle mapping feature of the polar coordinate deviation normalization algorithm is particularly well-suited for handling the complex relationships among multidimensional geological parameters in slope engineering. Its mathematical transformation mechanism effectively preserves the geometric characteristics between parameters, a property of great significance in slope stability analysis. The k-nearest neighbor interpolation mechanism of the synthetic minority oversampling algorithm is specifically designed to address the scarcity of slope instability samples. Its linear interpolation strategy rationally generates new samples within the feature space of the original instability samples, avoiding the non-physical significance of random generation. The second-order convergence property of the Newton-Raphson optimization algorithm enables rapid localization of the global optimal solution in the complex parameter space of slope stability discriminant models. Compared with traditional gradient descent methods, it exhibits significant advantages in handling non-convex optimization problems in extreme gradient boosting models. The multi-fold evaluation mechanism of ten-fold cross-validation ensures the robustness of the slope early warning model under different geological conditions and monitoring scenarios. Its statistical properties are particularly suitable for data-limited and complex environments in slope engineering applications. The game-theoretic basis of the Shapley value calculation provides a theoretically rigorous method for quantifying feature importance for slope stability analysis. Its fair distribution principle accurately reflects the true contribution of each geological parameter to slope instability. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 This is a schematic diagram of an embodiment of a slope instability early warning method in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a slope instability early warning system in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of the slope instability early warning device in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a slope instability early warning method and system. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0015] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the slope instability early warning method of the present application, the method includes: Step S101: normalize the original slope monitoring data by polar coordinate deviation normalization to obtain a standardized slope characteristic vector; Step S102, expanding the minority class samples in the standardized slope feature vector according to the synthetic minority class oversampling to obtain a slope training data set; Step S103, performing iterative optimization processing on the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model using Newton-Raphson optimization to obtain a slope stability discrimination model; Step S104: Calculate the accuracy and precision of the slope training data set and the slope stability discrimination model through ten-fold cross validation to obtain the slope stability discrimination model; Step S105: Perform weighted analytical processing on the gravity density, cohesion, and internal friction angle features in the slope stability discrimination model based on interpretability analysis to obtain a slope instability risk warning output.

[0016] It is understandable that the execution subject of the present application can be a slope instability warning system, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking a server as the execution subject as an example.

[0017] Specifically, the collected parameters such as gravity density, cohesion, internal friction angle, slope height, and slope gradient are numerically extracted to generate a slope characteristic parameter matrix. Each parameter value is then converted into the corresponding angular and radial coordinates through polar coordinate transformation. Polar coordinate transformation is to convert the data points in the rectangular coordinate system into a coordinate system represented by angle and distance from the origin. The deviation of each parameter relative to the mean in the polar coordinate system is then calculated, and the deviation is calculated using mean subtraction and standard deviation division. Finally, all parameter values ​​are mapped to the same numerical range through standardization and normalization. After completing the vectorized reconstruction, the standardized slope characteristic vector is obtained.

[0018] The standardized slope feature vector is used to identify sample category labels, and the number distribution of stable slope samples and unstable slope samples is statistically analyzed. Usually, unstable samples are far less than stable samples. Then, the k-nearest neighbor algorithm is used to search the neighborhood of the minority unstable samples. The k-nearest neighbor algorithm finds the k closest neighbor samples of each unstable sample by calculating the Euclidean distance between samples. Then, a linear interpolation algorithm is used to generate new synthetic samples between the unstable sample and its neighbor samples. Linear interpolation is a mathematical method that generates intermediate values ​​between two known data points according to a certain ratio. Finally, the synthetic unstable samples are merged and rearranged with the original stable samples to generate a category-balanced slope training dataset.

[0019] The Newton-Raphson optimization algorithm is used to optimize the hyperparameters of the extreme gradient boosting model. The extreme gradient boosting model is a machine learning algorithm based on the gradient boosting decision tree. It requires setting key parameters such as maximum depth, learning rate, subsampling rate, column sampling rate, and minimum loss. The Newton-Raphson optimization algorithm is a second-order optimization method. It determines the parameter update direction by calculating the first-order gradient and second-order Hessian matrix of the objective function. The Hessian matrix is ​​a square matrix containing all second-order partial derivatives of the objective function and can provide more accurate curvature information. The algorithm calculates the direction vector of the parameter update based on the gradient and the Hessian matrix, and then adjusts the step size and updates the values ​​of the initial hyperparameter combination. The optimized parameters are input into the extreme gradient boosting model for training. After convergence judgment and parameter solidification, the slope stability discrimination model is obtained.

[0020] The model performance was evaluated through ten-fold cross-validation. Ten-fold cross-validation is to randomly divide the data set into ten equal subsets, using nine subsets as training data each time and the remaining subset as test data. This is repeated ten times to ensure that each subset is used as a test set once. The trained slope discrimination model predicts and classifies the test set, and compares the prediction results with the true labels to calculate performance indicators such as accuracy, precision, recall, and F1 score. Accuracy is the proportion of correctly predicted samples to the total number of samples, precision is the proportion of samples predicted to be unstable and actually unstable to all predicted unstable samples, recall is the proportion of samples predicted to be unstable and actually unstable to all actually unstable samples, and F1 score is the harmonic mean of precision and recall. Finally, the performance indicators of the ten verifications are averaged and the variance is calculated to obtain the final slope stability discrimination model.

[0021] The interpretability analysis method is used to analyze the model feature weights, and the importance analysis of features such as gravity density, cohesion, internal friction angle, slope height, and slope is performed based on the Shapley value calculation method. The Shapley value is derived from cooperative game theory and is used to measure the marginal contribution of each feature to the model prediction results. By enumerating all possible feature combinations and calculating the prediction performance difference under each combination, the marginal contribution of each feature under different combinations is weighted averaged to obtain the Shapley value of a single feature. The Shapley value is normalized to obtain the standardized feature importance score. The features are arranged in descending order and graded according to the importance score to generate a sequence of key influencing factors for slope instability. Finally, the influencing factor sequence is compared with the preset risk threshold to determine the risk level. The corresponding warning signal strength and time window parameters are configured to generate a formatted slope instability risk warning output.

[0022] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The gravity density, cohesion, internal friction angle, slope height and slope parameters in the original data of slope monitoring are numerically extracted and processed to obtain the slope characteristic parameter matrix; Based on polar coordinate transformation, each parameter in the slope characteristic parameter matrix is ​​subjected to angle mapping processing to obtain polar coordinate slope characteristic data; The deviation of the polar coordinate slope characteristic data is calculated and processed by the mean subtraction and standard deviation division to obtain the deviation statistics; The polar coordinate slope characteristic data and the deviation statistics are normalized to obtain normalized slope characteristic data; The normalized slope characteristic data is vectorized and reconstructed to obtain the standardized slope characteristic vector.

[0023] Specifically, when performing numerical extraction and processing on the raw data of slope monitoring, it is necessary to identify and separate key geomechanical parameters from the multi-source heterogeneous data collected by the slope monitoring equipment. The gravity density is obtained by measuring the unit volume mass of the slope soil with a densitometer. The cohesion is determined by measuring the cohesive strength between soil particles through direct shear tests or triaxial tests. The internal friction angle is calculated by the Mohr circle analysis method as the ratio of the internal friction resistance of the soil to the normal stress. The slope height is measured by a laser rangefinder as the vertical distance from the top of the slope to the foot of the slope. The slope is measured by an inclinometer as the angle between the slope surface and the horizontal plane. These parameters with different dimensions and numerical ranges are arranged in rows to generate a slope characteristic parameter matrix. Each row in the matrix represents a complete parameter combination at a monitoring moment, and each column corresponds to a specific geological parameter type.

[0024] The angle mapping processing of the slope characteristic parameter matrix based on polar coordinate transformation is to convert the parameter values ​​in the rectangular coordinate system into the polar coordinate system. Polar coordinate transformation is a mathematical coordinate conversion method. Each parameter value is used as the radial distance. The corresponding angle coordinate is assigned according to the position of the parameter in the matrix. The gravity density corresponds to the zero-degree reference angle, the cohesion corresponds to the seventy-two-degree angle, the internal friction angle corresponds to the one hundred and forty-four-degree angle, the slope height corresponds to the two hundred and sixteen-degree angle, and the slope corresponds to the two hundred and eighty-eight-degree angle. Through angle mapping, different types of physical quantities are uniformly mapped to the same polar coordinate space. Each parameter value corresponds to a polar coordinate point composed of the angle and radial distance to generate polar coordinate slope characteristic data.

[0025] Deviation calculation of polar coordinate slope characteristic data based on mean subtraction and standard deviation division is the core step of standardization preprocessing. Mean subtraction is to subtract the historical mean of the parameter type from the radial distance of each polar coordinate data point to eliminate the numerical benchmark differences between different parameters. Standard deviation division is to divide the result after mean subtraction by the historical standard deviation of the parameter type to eliminate the numerical distribution differences between different parameters. The standard deviation is a statistic that measures the degree of data dispersion. It is obtained by calculating the square root of the sum of the squares of the differences between the radial distance and the mean of each parameter. The deviation calculation process converts parameters with different numerical ranges into a standard normal distribution with a mean of zero and a standard deviation of one to obtain the deviation statistic.

[0026] Standardizing and normalizing the polar coordinate slope characteristic data and deviation statistics is a key step in further unifying the data scale. Standardization is a linear transformation of the polar coordinate data based on the deviation statistics. Normalization is to map the standardized values ​​to a fixed interval from zero to one. The specific operation is to add the minimum absolute value of the deviation statistics to each standardized polar coordinate radial distance, and then divide it by the difference between the maximum and minimum values ​​of the deviation statistics. The normalization process eliminates the influence of different dimensions and numerical ranges between parameters, so that all parameters have the same weight basis, and obtain normalized slope characteristic data with a unified numerical range.

[0027] Vectorized reconstruction of normalized slope characteristic data is to reorganize the scattered polar coordinate data points into a data structure suitable for machine learning algorithm input. Vectorized reconstruction arranges the five normalized parameters of each monitoring moment into a one-dimensional vector in a fixed order of gravity density, cohesion, internal friction angle, slope height, and slope. Each vector contains five elements, corresponding to the normalized values ​​of the five geological parameters. The vector combination of multiple monitoring moments generates a two-dimensional standardized slope characteristic vector matrix. The number of matrix rows is equal to the number of monitoring moments, and the number of columns is fixed to five. Vectorized reconstruction ensures that the data format fully matches the input requirements of the subsequent machine learning algorithm.

[0028] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Perform sample category label recognition and statistical counting processing on the standardized slope feature vector to obtain majority class samples and minority class samples; Based on the k-nearest neighbor algorithm, the neighborhood search and distance calculation processing of the minority class samples are performed to obtain the neighborhood set of the minority class samples; According to the linear interpolation algorithm, the minority class samples and the minority class sample neighborhood set are randomly combined and interpolated to obtain synthetic minority class samples; The synthesized minority class samples and majority class samples are merged and rearranged to obtain class-balanced slope sample data; The class-balanced slope sample data are divided into training set and validation set to obtain the slope training data set.

[0029] Specifically, to perform sample category label identification and statistical counting processing on the standardized slope feature vector, it is first necessary to assign corresponding category labels to each standardized feature vector based on the historical monitoring records of the slope and actual instability events. The stable slope is marked as category zero and the unstable slope is marked as category one. The category labels come from the comprehensive judgment of geological survey reports, slope deformation monitoring data and actual landslide occurrence records. The statistical counting processing calculates the number distribution of zero-category labels and category one labels by traversing the category labels of all samples. The majority class samples refer to the sample category with a larger number, usually stable slope samples, and the minority class samples refer to the sample category with a smaller number, usually unstable slope samples. Since slope instability events are relatively rare, the number of unstable samples is far less than the number of stable samples, resulting in a serious category imbalance problem. The statistical results directly affect the number of samples generated by subsequent synthetic oversampling. The core of synthetic minority oversampling is to perform neighborhood search and distance calculation on minority class samples based on the k-nearest neighbor algorithm. The k-nearest neighbor algorithm is an instance-based learning method that finds the closest neighbor samples by calculating the similarity between samples. The distance calculation uses the Euclidean distance formula, and the square root of the sum of the squares of the differences between the corresponding elements of the two sample vectors is used to obtain the spatial distance between samples. The neighborhood search process is to calculate the Euclidean distance between each unstable minority class sample and all other minority class samples, and select the first k nearest neighbor samples in ascending order of distance. The k value is usually set to five to ensure that the neighborhood contains sufficient sample information while avoiding noise interference. The minority class sample neighborhood set contains each unstable sample and its corresponding k nearest neighbor samples. The construction of the neighborhood set establishes a spatial foundation for the subsequent linear interpolation synthesis of new samples.

[0030] Randomly combining and interpolating minority samples and their neighborhood sets according to the linear interpolation algorithm is a key step in generating synthetic samples. The linear interpolation algorithm is a mathematical method that generates intermediate values ​​between two known data points in a certain proportion. Random combination is to randomly select a neighbor sample from the neighborhood set of each minority sample for pairing. The interpolation calculation performs a linear combination between the original minority sample and the randomly selected neighbor sample. The specific calculation is to multiply the original sample vector by a randomly generated weight coefficient between zero and one, and the neighbor sample vector by one minus the weight coefficient. The two result vectors are added together to obtain the synthetic minority sample. The randomness of the weight coefficient ensures that the synthetic samples are distributed at different positions on the line connecting the original sample and the neighbor sample. The synthetic samples generated by interpolation inherit the characteristic pattern of the original unstable sample while increasing the diversity and quantity of the samples.

[0031] Merging and rearranging the data of the synthetic minority class samples and the majority class samples is an integration step to balance the sample distribution. Data merging is to combine the newly generated synthetic unstable samples with the original stable samples and unstable samples to generate an extended sample set. The rearrangement process eliminates the temporal sequence and regularity of the data by randomly disrupting the order of samples, avoiding batch bias in the training process. Class-balanced slope sample data refers to a data set in which the number of unstable samples is basically equal to the number of stable samples. The balancing process eliminates the negative impact of class imbalance on model training, so that the model has the same learning emphasis on the minority class unstable samples and the majority class stable samples.

[0032] Dividing the class-balanced slope sample data into training sets and validation sets is a key step in allocating data for model training and evaluation. Training set division is to select samples from the class-balanced samples in a certain proportion for model parameter learning, and validation set division is to select some samples from the remaining samples for model performance evaluation and parameter tuning. The division ratio usually adopts an eight-to-two distribution method, that is, eighty percent of the samples are used as training sets and twenty percent of the samples are used as validation sets. The division process needs to keep the ratio of unstable samples to stable samples in the training set and validation set consistent. The slope training dataset contains balanced training samples and validation samples, and has the data foundation required to support machine learning model training.

[0033] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Initialize the maximum depth, learning rate, subsampling rate, column sampling rate, and minimum loss parameters of the extreme gradient boosting model to obtain the initial hyperparameter combination; Based on the Newton-Raphson iterative algorithm, the gradient calculation and Hessian matrix calculation of the initial hyperparameter combination are performed to obtain the parameter update direction vector; According to the parameter update direction vector, the initial hyperparameter combination is adjusted in step size and updated in value to obtain the optimized hyperparameter combination; The optimized hyperparameter combination is input into the extreme gradient boosting model for model construction and training to obtain a trained slope discrimination model; The trained slope discrimination model is subjected to convergence judgment and parameter solidification processing to obtain a slope stability discrimination model.

[0034] Specifically, initializing the maximum depth, learning rate, subsampling rate, column sampling rate, and minimum loss parameters of the extreme gradient boosting model is the starting point of hyperparameter optimization. The maximum depth parameter controls the maximum number of layers of the decision tree and determines the complexity and fitting ability of the model. The initial value is usually set to six layers. The learning rate parameter controls the step size of the gradient update in each iteration, affecting the model convergence speed and final performance. The initial value is set to 0.1. The subsampling rate parameter controls the proportion of randomly selected samples during each training to prevent overfitting and increase the generalization ability of the model. The initial value is set to 0.8. The column sampling rate parameter controls the proportion of randomly selected features each time a decision tree is constructed to increase the randomness and robustness of the model. The initial value is set to 0.6. The minimum loss parameter controls the minimum loss reduction required when a leaf node is split to avoid over-splitting and overfitting. The initial value is set to 0.1. The combination of these five parameters generates the initial hyperparameter combination, which directly affects the training effect and prediction performance of the extreme gradient boosting model. The core calculation steps of parameter optimization are to calculate the gradient and Hessian matrix of the initial hyperparameter combination based on the Newton-Raphson iterative algorithm. The Newton-Raphson iterative algorithm is a second-order optimization method that finds the optimal solution by simultaneously utilizing the first-order derivative and second-order derivative information of the objective function. The gradient calculation is to solve the first-order partial derivative of the objective function with respect to each hyperparameter, reflecting the direction and rate of change of the objective function at the current parameter point. The Hessian matrix calculation is to solve the second-order partial derivative of the objective function with respect to each hyperparameter and form a symmetric square matrix, reflecting the curvature information of the objective function at the current parameter point. The product of the inverse matrix of the Hessian matrix and the gradient vector obtains the parameter update direction vector, which indicates the parameter adjustment direction that minimizes the objective function value. Compared with the method that only uses the first-order gradient, the Newton-Raphson algorithm takes into account the curvature characteristics of the function, and converges faster and with higher accuracy.

[0035] The execution stage of parameter optimization is to adjust the step size and update the numerical value of the initial hyperparameter combination according to the parameter update direction vector. The step size adjustment is to determine the appropriate moving distance in the parameter update direction to avoid oscillation caused by excessive parameter update or slow convergence caused by excessive parameter update. The step size is determined by the line search method. Different step size values ​​are tried in the update direction and the step size that minimizes the objective function value is selected. The numerical update processing is to add the current hyperparameter value to the product of the step size and the update direction vector to obtain a new hyperparameter value. The updated maximum depth, learning rate, subsampling rate, column sampling rate, and minimum loss parameters generate an optimized hyperparameter combination. This combination is closer to the optimal point of the objective function than the initial setting. The iterative process is repeated until the parameter change is less than the preset threshold or the maximum number of iterations is reached.

[0036] Inputting the optimized hyperparameter combination into the extreme gradient boosting model for model construction and training is a key step in converting the optimization results into an actual model. Model construction is to create the structural framework of the extreme gradient boosting model based on the optimized hyperparameter configuration, including the depth limit of the decision tree, learning rate setting, sampling strategy configuration, etc. Training is to use the slope training data set to learn the parameters of the model. The extreme gradient boosting algorithm adopts the ensemble learning method of gradient boosting decision trees. By iteratively constructing multiple decision trees and combining their prediction results, each decision tree is fitted to the residual of the previous round of prediction, gradually reducing the prediction error. During the training process, the model learns the nonlinear mapping relationship between slope characteristics and instability risk, and automatically captures the complex interaction of characteristics such as gravity density, cohesion, and internal friction angle. The trained slope discrimination model has the ability to classify the stability of new slope samples.

[0037] Convergence judgment and parameter solidification of the trained slope discrimination model are the final steps of model optimization. Convergence judgment is the inspection process to evaluate whether the model training has reached a stable state. The convergence state is determined by monitoring the training loss function and verifying the changing trend of the loss function. When the change of the loss function in multiple consecutive iterations is less than the preset threshold, it is judged to be converged. Parameter solidification is to save the converged model parameters in an immutable state, including the node splitting rules, leaf node prediction values, feature weights, etc. of all decision trees. The solidified parameters no longer change with the changes in training data. The slope stability discrimination model becomes the final model with deterministic prediction capabilities. The model contains a complete mapping function from slope characteristics to instability probability.

[0038] In a specific embodiment, the process of executing step S104 may specifically include the following steps: The slope training data set is randomly divided into ten equal parts to obtain ten sub-data sets; Nine of the ten sub-datasets are used as training sets to input the slope stability discrimination model for training, thereby obtaining a trained slope discrimination model. Based on the trained slope discrimination model, the remaining sub-dataset is subjected to prediction and classification processing to obtain the slope stability prediction result; The accuracy, precision, recall, and F1 score are calculated based on the slope stability prediction results and the true labels to obtain the single verification performance indicators; The single verification performance index was averaged and the variance was calculated ten times to obtain the slope stability discrimination model.

[0039] Specifically, all samples in the slope training dataset are randomly sorted to disrupt the original order to eliminate the temporal sequence and batch effect of the data. Then, the total number of samples is divided by ten to obtain the target capacity of each sub-dataset. Random division is to assign continuous sample indexes to different sub-datasets by equally spaced sampling to ensure that each sub-dataset contains the same number of samples. At the same time, the proportion of stable slope samples and unstable slope samples in each sub-dataset is kept consistent. The ten sub-datasets are marked as the first to the tenth folds, and each sub-dataset contains a feature vector and a corresponding category label. The divided sub-datasets take turns to serve as the test set in the subsequent verification process, and the remaining nine sub-datasets are used as training sets.

[0040] The core of the cross-validation cycle is to input nine of the ten sub-datasets as training sets into the slope stability discrimination model for training processing. At the beginning of the training process, the first sub-dataset is selected as the test set, and the remaining nine sub-datasets are merged row by row to generate a temporary training set. The temporary training set contains 90% of the sample information of the original training data. The input processing is to pass the feature vector and category label of the temporary training set to the input interface of the slope stability discrimination model respectively. The model re-learns the mapping relationship between features and labels based on the input training samples. During the training process, the model adjusts the internal decision tree structure, node splitting threshold and leaf node weight. The learning process adopts the iterative optimization method of gradient boosting. Each iteration constructs a new decision tree based on the prediction residual of the previous round. After the training is completed, a trained slope discrimination model optimized for the current 9-fold data is obtained.

[0041] The key step in evaluating model performance is to perform predictive classification processing on the remaining sub-dataset based on the trained slope discrimination model. The predictive classification processing inputs the five-dimensional feature vector of each sample in the test set into the trained slope discrimination model. Multiple decision trees within the model judge and classify the input features respectively. Each decision tree reaches the leaf node along the branch path based on the comparison result of the feature value and the node splitting threshold. The leaf node outputs the probability value of the sample belonging to the unstable category. The probability output of all decision trees is weighted averaged to obtain the final instability probability. When the instability probability is greater than 0.5, it is predicted to be an unstable slope. When it is less than or equal to 0.5, it is predicted to be a stable slope. The prediction result contains the predicted category label and corresponding confidence probability of each test sample. The slope stability prediction result is compared with the true label in the test set.

[0042] Calculating the accuracy, precision, recall and F1 score based on the slope stability prediction results and the true labels is a statistical step for quantifying the model performance. The accuracy calculation is the number of correctly predicted samples divided by the total number of test samples, reflecting the accuracy of the model's overall classification. The precision calculation is the number of samples predicted to be unstable and actually unstable divided by the number of all predicted unstable samples, reflecting the reliability of the instability prediction. The recall calculation is the number of samples predicted to be unstable and actually unstable divided by the number of all actually unstable samples, reflecting the model's ability to identify unstable samples. The F1 score calculation is the harmonic mean of the precision and recall, which comprehensively reflects the model's balanced performance in the instability detection task. The harmonic mean is the product of twice the precision and recall divided by the sum of the precision and recall. The single verification performance index includes the numerical results of the four evaluation dimensions.

[0043] The ten-cycle averaging and variance calculation processing of the single validation performance index is a comprehensive stage of cross-validation result statistics. Ten cycles refer to repeating the aforementioned training and testing process ten times, each time selecting a different sub-dataset as the test set, and the remaining nine sub-datasets as the training set. After the cycle, ten sets of single validation performance indicators are obtained. The average calculation processing is to sum up the ten accuracy rates, precision rates, recall rates, and F1 scores and divide them by ten to obtain the average value of each performance indicator. The variance calculation processing is to square the difference between each indicator value and the corresponding average value, then sum it and divide it by nine, and then take the square root to obtain the standard deviation. The average value reflects the overall performance level of the model, and the variance reflects the stability of the model performance. A small variance indicates that the model has consistent performance under different data partitions, and a large variance indicates that the model is sensitive to the training data. The final slope stability discrimination model contains performance evaluation results and model stability indicators.

[0044] In a specific embodiment, the process of executing step S105 may specifically include the following steps: Based on the Shapley value calculation, the marginal contribution of gravity density, cohesion, internal friction angle, slope height and slope gradient in the slope stability judgment model is calculated and processed to obtain the feature importance value; According to the feature importance values, each feature is arranged in descending order and divided into importance levels to obtain the sequence of key influencing factors of slope instability; Compare and analyze the sequence of key factors affecting slope instability with the preset risk threshold and determine the risk level to obtain the slope instability risk level; Based on the slope instability risk level, the warning signal strength and warning time window are parameterized and processed to obtain the slope instability warning parameters; According to the slope instability warning parameters, warning information is generated and output formatted to obtain the slope instability risk warning output.

[0045] Specifically, the marginal contribution calculation of gravity density, cohesion, internal friction angle, slope height and slope gradient in the slope stability discrimination model based on Shapley value calculation is the core step of interpretability analysis. Shapley value comes from cooperative game theory and is used to fairly distribute the contribution value of each member in the alliance. The marginal contribution calculation process first enumerates all possible combinations of the five features, including single feature, two feature combination, three feature combination, four feature combination and all feature combinations, totaling 31 different feature subsets. For each feature subset, the slope stability discrimination is input. The model performs prediction performance evaluation, records the prediction accuracy of each subset as the contribution value of the subset, and then calculates the marginal contribution of each feature in different subsets. The marginal contribution is the difference between the prediction performance of the subset containing the feature and the prediction performance of the subset excluding the feature. The Shapley value calculation is a weighted average of the marginal contribution of each feature in all possible subsets. The weight is determined by the factorial of the subset size and the factorial of the number of remaining features. Finally, a fair contribution measure of each feature to the overall performance of the model is obtained. The feature importance value reflects the actual impact of each feature on the prediction of slope instability.

[0046] Arranging the features in descending order and dividing their importance levels according to their importance values ​​are the sorting and classification steps of feature analysis. The descending order sorts the five features of gravity density, cohesion, internal friction angle, slope height, and slope from large to small according to their Shapley values. The larger the Shapley value, the more significant the contribution of the feature to instability prediction. The arrangement result generates a priority order of feature importance. The importance level classification process divides the features into three levels of high importance, medium importance, and low importance according to the numerical range of the Shapley value. The Shapley value of the high-importance feature is greater than 0.3, the Shapley value of the medium-importance feature is between 0.1 and 0.3, and the Shapley value of the low-importance feature is less than 0.1. The level classification results directly affect the formulation of early warning strategies and the allocation of monitoring priorities. The sequence of key influencing factors of slope instability is arranged from high to low according to importance, which lays a scientific basis for subsequent risk assessment and early warning decisions.

[0047] Comparative analysis of the sequence of key influencing factors of slope instability with preset risk thresholds and risk level determination are key links in risk quantification. The preset risk thresholds are determined based on statistical analysis of historical slope instability events and expert experience, including three level standards: high risk threshold, medium risk threshold, and low risk threshold. The comparative analysis compares the actual monitoring values ​​of each feature of the current slope with the corresponding risk threshold. When a high-importance feature exceeds the high-risk threshold, it is determined to be in a high-risk state. When a medium-importance feature exceeds the medium-risk threshold, it is determined to be in a medium-risk state. When a low-importance feature exceeds the low-risk threshold, it is determined to be in a low-risk state. The risk level determination adopts the highest risk principle, that is, when multiple features exceed different level thresholds at the same time, the highest risk level is used as the final determination result. The slope instability risk level is divided into four levels, including safety level, attention level, warning level, and danger level. The risk level directly determines the urgency of the early warning response and the strictness of the disposal measures.

[0048] Parameter configuration of warning signal strength and warning time window based on slope instability risk level is a step in formulating warning strategy. Warning signal strength configuration sets different signal output power and propagation range according to risk level. No warning signal is sent at safety level, low-intensity signal is sent at attention level to cover monitoring personnel, medium-intensity signal is sent at alert level to cover management department, and high-intensity signal is sent at danger level to cover all relevant personnel. Warning time window configuration refers to the time interval setting from the discovery of risk to the expected occurrence of instability. The time window for attention level is seventy-two hours, the time window for alert level is twenty-four hours, and the time window for danger level is six hours. Parameter configuration processing also includes warning frequency setting. Attention level, warning is once a day, warning level, once every four hours, and danger level, once an hour. Slope instability warning parameters integrate configuration information of three dimensions: signal strength, time window, and warning frequency.

[0049] The generation and output formatting of warning information based on slope instability warning parameters is the final output link of the warning system. The warning information generation and processing constructs structured warning content according to the current risk level and key influencing factor sequence, including key information such as risk level identification, main risk characteristics, expected instability time, and recommended disposal measures. The output formatting processing encodes and encapsulates the warning information in a standardized format, supporting multiple transmission channels such as SMS, email, web pages, and mobile applications. The formatted content includes structured data such as timestamp, geographic coordinates, risk level code, influencing factor ranking, and confidence assessment. The slope instability risk warning output adopts a unified data interface and communication protocol to ensure compatibility and interoperability with existing monitoring systems and emergency response systems.

[0050] In a specific embodiment, the step of calculating the marginal contribution of gravity density, cohesion, internal friction angle, slope height, and slope gradient characteristics in the slope stability judgment model based on Shapley value calculation may specifically include the following steps: The gravity density, cohesion, internal friction angle, slope height and slope gradient in the slope stability discrimination model are processed by feature subset enumeration and combination generation to obtain a feature combination set. Based on the alliance game theory, the prediction performance difference of each feature combination in the feature combination set is calculated to obtain the feature marginal contribution value; Based on the marginal contribution value of the feature, the weighted average and Shapley value calculation of each feature under different feature combinations are performed to obtain the Shapley value of a single feature; Normalize the single feature Shapley value and convert its relative importance to obtain the standardized feature importance score; The standardized feature importance scores are numerically ranked and importance quantified to obtain the feature importance values.

[0051] Specifically, the feature subset enumeration and combination generation processing of gravity density, cohesion, internal friction angle, slope height and slope gradient in the slope stability discrimination model are the basic steps for Shapley value calculation. Feature subset enumeration is based on the principle of combinatorial mathematics to systematically list all possible combinations of five features. The enumeration process is carried out from zero to five according to the number of features contained in the subset. Zero features constitute the empty set, one feature constitutes five single-element subsets, two features constitute ten two-element subsets, three features constitute ten three-element subsets, four features constitute five four-element subsets, and five features constitute a full set. The combination generation processing adopts the binary bit operation method, corresponding to each feature to a binary bit, and the inclusion and exclusion status of the feature is represented by the change of the bit value from zero to one. The binary representation of integers from zero to thirty-one corresponds to thirty-two different feature combinations. The feature combination set contains all possible feature subsets from the empty set to the full set, laying a combination foundation for the subsequent marginal contribution calculation. Calculating the prediction performance difference of each feature combination in the feature combination set according to alliance game theory is the core link of quantifying feature contribution. Alliance game theory is a mathematical theory that studies the value distribution generated by the cooperation of multiple participants. In slope instability warning, each feature is regarded as a game participant, and the feature combination is regarded as a participant alliance. The prediction performance difference calculation is to subtract the prediction accuracy of the alliance containing the target feature from the prediction accuracy of the alliance excluding the target feature. The difference reflects the performance improvement brought about by the target feature joining the alliance. The calculation process requires calculating the marginal contribution of each feature in all alliances containing the feature separately. The marginal contribution refers to the impact of adding or removing a feature on the model performance under the premise of fixing other feature combinations. The positive and negative and size of the feature marginal contribution value directly reflect the promotion or obstruction of the feature on slope instability prediction and its intensity.

[0052] Based on the marginal contribution value of the feature, weighted averaging and Shapley value calculation of each feature under different feature combinations is a mathematical method for fairly distributing feature value. The weighted averaging process takes into account the impact of alliance size on marginal contribution. The marginal contribution in a smaller alliance is given a higher weight, and the marginal contribution in a larger alliance is given a lower weight. The weight calculation is based on the factorial formula in combinatorics. The factorial of the number of leading features is multiplied by the factorial of the number of trailing features and then divided by the factorial of the total number of features. The Shapley value calculation is to sum the weighted marginal contributions of a feature in all possible alliances. The calculation result satisfies the four axiomatic properties of efficiency, symmetry, virtuality, and additivity. Efficiency ensures that the sum of the Shapley values ​​of all features is equal to the total value of the entire alliance. Symmetry ensures that features with the same contribution to the model obtain the same Shapley value. Virtuality ensures that features that do not contribute to the model obtain zero Shapley value. Additivity ensures that the Shapley values ​​of multiple independent games are equal to the sum of the Shapley values ​​of each game. The Shapley value of a single feature represents the fair contribution of the feature to the slope instability prediction model.

[0053] Normalizing the Shapley values ​​of individual features and performing relative importance conversion processing are numerical transformation steps for standardized feature values. Normalization processing is to divide the Shapley values ​​of all features by the sum of their Shapley values ​​to ensure that the sum of the normalized values ​​is one, eliminating the influence of the absolute value size and highlighting the relative importance differences between features. Relative importance conversion is to multiply the normalized value by 100 to convert it into a percentage form, which is convenient for intuitive understanding and comparative analysis. The conversion process also includes sign processing. A positive value indicates that the feature has a positive contribution to the instability prediction, and a negative value indicates that the feature has a negative impact on the instability prediction. The absolute value size indicates the contribution strength. The standardized feature importance score eliminates the dimension effect and the numerical scale difference, and generates a unified feature value measurement standard.

[0054] Numerical sorting and importance quantification of standardized feature importance scores are the final steps in establishing feature priorities. The numerical sorting process arranges the five features in descending order according to the absolute value of the standardized score. The sorting result directly reflects the importance level of each feature to slope instability prediction. The importance quantification process divides the features into different importance levels according to the score value range. The standardized score of high-importance features is greater than 0.3, corresponding to 30%, the score of medium-importance features is between 0.1 and 0.3, and the score of low-importance features is less than 0.1. The quantitative level provides a scientific basis for the subsequent monitoring strategy formulation and early warning parameter configuration. The feature importance value integrates information from two dimensions: sorting position and quantitative level, and generates a feature value evaluation system.

[0055] The above describes the slope instability warning method in the embodiment of the present application. The following describes the slope instability warning system in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the slope instability early warning system includes: A processing module is used to normalize the original data of slope monitoring by polar coordinate deviation normalization to obtain a standardized slope characteristic vector; An expansion module, configured to expand minority class samples in the standardized slope feature vector according to synthetic minority class oversampling to obtain a slope training data set; The optimization module is used to iteratively optimize the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model using Newton-Raphson optimization to obtain a slope stability discrimination model; a calculation module, configured to calculate the accuracy and precision of the slope training data set and the slope stability discrimination model through ten-fold cross validation to obtain the slope stability discrimination model; The analytical module is used to perform weighted analytical processing on the gravity density, cohesion, and internal friction angle characteristics in the slope stability discrimination model based on interpretability analysis to obtain a slope instability risk warning output.

[0056] above Figure 2 The slope instability warning system in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The slope instability warning device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0057] Reference Figure 3 In an embodiment of the present invention, a slope instability warning device is also provided. The slope instability warning device can be a server, and its internal structure can be as follows: Figure 3 As shown. The slope instability warning device includes a processor, a memory, a display screen, an input device, a network interface and a database connected via a system bus. The computer-designed processor is used to provide computing and control capabilities. The memory of the slope instability warning device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the slope instability warning device is used to store the corresponding data in this embodiment. The network interface of the slope instability warning device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0058] Those skilled in the art will understand that Figure 3The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the slope instability early warning device to which the solution of the present invention is applied.

[0059] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, cause the computer to execute the steps of the slope instability warning method.

[0060] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0061] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a slope instability warning device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0062] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A slope instability early warning method, characterized in that: The method comprises: The original data of slope monitoring are normalized by polar coordinate deviation standardization to obtain the standardized slope characteristic vector; performing expansion processing on minority class samples in the standardized slope feature vector according to synthetic minority class oversampling to obtain a slope training data set; The Newton-Raphson optimization was used to iteratively optimize the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model to obtain a slope stability discrimination model. The slope training data set and the slope stability discrimination model are processed for accuracy and precision through ten-fold cross validation to obtain a slope stability discrimination model; According to the interpretability analysis, the gravity density, cohesion and internal friction angle characteristics in the slope stability discrimination model are weighted and processed to obtain the slope instability risk warning output.

2. The slope instability early warning method according to claim 1, characterized in that: The normalization of the slope monitoring raw data by polar coordinate deviation normalization to obtain a standardized slope characteristic vector includes: The gravity density, cohesion, internal friction angle, slope height and slope parameters in the original data of slope monitoring are numerically extracted and processed to obtain the slope characteristic parameter matrix; Performing angle mapping processing on each parameter in the slope characteristic parameter matrix based on polar coordinate transformation to obtain polar coordinate slope characteristic data; performing deviation calculation processing on the polar coordinate slope characteristic data according to mean subtraction and standard deviation division to obtain deviation statistics; performing normalization processing on the polar coordinate slope characteristic data and the deviation statistics to obtain normalized slope characteristic data; The normalized slope characteristic data is subjected to vectorized reconstruction processing to obtain a standardized slope characteristic vector.

3. The slope instability early warning method according to claim 1, characterized in that: The method of expanding the minority class samples in the standardized slope feature vector according to the synthetic minority class oversampling to obtain a slope training data set includes: Performing sample category label recognition and statistical counting processing on the standardized slope feature vector to obtain majority class samples and minority class samples; Performing neighborhood search and distance calculation on the minority class samples based on the k-nearest neighbor algorithm to obtain a neighborhood set of minority class samples; Performing random combination and interpolation calculation processing on the minority class samples and the minority class sample neighborhood set according to a linear interpolation algorithm to obtain a synthetic minority class sample; Merging and rearranging the synthesized minority class samples and the majority class samples to obtain class-balanced slope sample data; The class-balanced slope sample data are divided into a training set and a validation set to obtain a slope training data set.

4. The slope instability early warning method according to claim 1, characterized in that: The Newton-Raphson optimization is used to iteratively optimize the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model to obtain a slope stability discrimination model, including: Initialize the maximum depth, learning rate, subsampling rate, column sampling rate, and minimum loss parameters of the extreme gradient boosting model to obtain the initial hyperparameter combination; Performing gradient calculation and Hessian matrix calculation on the initial hyperparameter combination based on the Newton-Raphson iterative algorithm to obtain a parameter update direction vector; Performing step size adjustment and value update processing on the initial hyperparameter combination according to the parameter update direction vector to obtain an optimized hyperparameter combination; Inputting the optimized hyperparameter combination into the extreme gradient boosting model for model construction and training to obtain a trained slope discrimination model; The trained slope discrimination model is subjected to convergence judgment and parameter solidification processing to obtain a slope stability discrimination model.

5. The slope instability early warning method according to claim 1, characterized in that: The slope stability discrimination model is obtained by performing accuracy and precision calculation on the slope training data set and the slope stability discrimination model through ten-fold cross validation, including: The slope training data set is randomly divided into ten equal parts to obtain ten sub-data sets; inputting nine of the ten sub-data sets as training sets into the slope stability discrimination model for training processing to obtain a trained slope discrimination model; Performing prediction and classification processing on the remaining sub-dataset based on the trained slope discrimination model to obtain a slope stability prediction result; Calculate the accuracy, precision, recall, and F1 score based on the slope stability prediction results and the true labels to obtain a single verification performance index; The single verification performance index is subjected to ten cycles of averaging and variance calculation processing to obtain a slope stability discrimination model.

6. The slope instability early warning method according to claim 1, characterized in that: The interpretability analysis is used to perform weighted analytical processing on the gravity density, cohesion, and internal friction angle features in the slope stability discrimination model to obtain a slope instability risk warning output, including: Based on the Shapley value calculation, marginal contribution calculation is performed on the gravity density, cohesion, internal friction angle, slope height and slope characteristics in the slope stability discrimination model to obtain the characteristic importance value; According to the feature importance values, each feature is arranged in descending order and divided into importance levels to obtain a sequence of key influencing factors of slope instability; Comparing and analyzing the sequence of key influencing factors of slope instability with a preset risk threshold and conducting risk level determination processing to obtain a slope instability risk level; Performing parameter configuration processing on the warning signal strength and the warning time window based on the slope instability risk level to obtain the slope instability warning parameters; Warning information is generated and output formatted according to the slope instability warning parameters to obtain a slope instability risk warning output.

7. The slope instability early warning method according to claim 6, characterized in that: The Shapley value-based calculation performs marginal contribution calculation processing on the gravity density, cohesion, internal friction angle, slope height, and slope characteristics in the slope stability judgment model to obtain feature importance values, including: Performing feature subset enumeration and combination generation processing on the gravity density, cohesion, internal friction angle, slope height, and slope gradient features in the slope stability discrimination model to obtain a feature combination set; Performing prediction performance difference calculation on each feature combination in the feature combination set according to alliance game theory to obtain a feature marginal contribution value; Based on the feature marginal contribution value, weighted average and Shapley value calculation are performed on each feature under different feature combinations to obtain a single feature Shapley value; Normalizing and converting the single feature Shapley value to relative importance to obtain a standardized feature importance score; The standardized feature importance scores are numerically sorted and importance quantified to obtain feature importance values.

8. A slope instability early warning system, characterized in that: For implementing the slope instability early warning method according to any one of claims 1 to 7, the slope instability early warning system comprises: A processing module is used to normalize the original data of slope monitoring by polar coordinate deviation normalization to obtain a standardized slope characteristic vector; An expansion module, configured to expand minority class samples in the standardized slope feature vector according to synthetic minority class oversampling to obtain a slope training data set; The optimization module is used to iteratively optimize the maximum depth, learning rate, and subsampling rate parameters of the extreme gradient boosting model using Newton-Raphson optimization to obtain a slope stability discrimination model; a calculation module, configured to calculate the accuracy and precision of the slope training data set and the slope stability discrimination model through ten-fold cross validation to obtain the slope stability discrimination model; The analytical module is used to perform weighted analytical processing on the gravity density, cohesion, and internal friction angle characteristics in the slope stability discrimination model based on interpretability analysis to obtain a slope instability risk warning output.

9. A slope instability warning device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the slope instability early warning method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to execute the slope instability early warning method according to any one of claims 1 to 7.

Images