Integrated landslide susceptibility evaluation method based on XGB reinforcement model

Abstract

The invention provides an integrated landslide susceptibility evaluation method based on an XGB reinforcement model, and belongs to the technical field of geological disaster prediction.The method comprises the steps that collected evaluation factor data of a landslide disaster research area is preprocessed, grid points are obtained, and non-landslide points are generated, so that the evaluation factor data of the landslide disaster research area are obtained; extracting an original data table of all evaluation factors corresponding to the landslide points; constructing an evaluation data set by calculating an information amount value of the evaluation factor, and dividing the evaluation data set into a training set and a verification set; a CatBoost-XGB integration model is constructed, training is performed by using the training set, verification is performed by using the verification set, and an optimal integration model is obtained; importing all grid point data in the region into the integrated model, carrying out landslide disaster susceptibility analysis, and outputting a landslide disaster susceptibility prediction map; the problems that an existing integrated landslide susceptibility evaluation method cannot effectively capture dynamic evolution characteristics of a geological process and is limited in applicability in a rapidly changing environment are solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of geological disaster prediction, and particularly relates to an integrated landslide susceptibility evaluation method based on an XGB reinforcement model. BACKGROUND

[0002] The susceptibility evaluation of landslide disasters still faces the dual challenges of data driving and mechanism integration under complex geological environments. Traditional methods rely on the isolated analysis of static environmental factors (such as slope and lithology), which fail to effectively capture the dynamic evolution characteristics of geological processes, such as the coupling effect of rainfall infiltration and fracture expansion, which is often ignored. Existing models are mostly based on the spatial distribution of historical landslide points for statistical modeling, but lack quantitative characterization of the spatio-temporal heterogeneity of potential triggering mechanisms (such as seismic disturbance and human activities), resulting in limited applicability of the prediction results in rapidly changing environments. In addition, the fusion application of multi-source remote sensing data (such as InSAR deformation and thermal infrared images) is still in the exploratory stage, and the data scale difference and noise interference make it difficult to extract high-precision features, which restricts the recognition ability of the model for hidden landslide hazards.

[0003] Although the current machine learning-driven evaluation method has made breakthroughs in accuracy, the model's transferability and robustness still have hidden problems. The generalization ability of a single model in different geomorphic units differs significantly, for example, a model trained in mountainous environments may have a significant performance decline when applied to plateau regions due to the introduction of special mechanisms such as freeze-thaw cycles or glacial action. At the same time, the black box model lacks feature explanation, making it difficult to support decision-making needs for disaster prevention and control - for example, the model may overestimate the contribution of a factor, but it cannot reveal the physical correlation between the factor and the landslide triggering threshold. Although ensemble learning techniques (such as Stacking and Boosting) improve prediction stability through model combination, most methods still remain at the level of "mechanical" weight superposition and fail to build a hierarchical feature-mechanism collaborative optimization framework, limiting their practical application value in multi-scale risk assessment.

[0004] The existing integrated landslide susceptibility evaluation method has limitations, such as insufficient feature selection and weight distribution, and low evaluation accuracy and robustness. SUMMARY

[0005] In view of the above shortcomings in the prior art, the integrated landslide susceptibility evaluation method based on the XGB reinforcement model provided by the present application solves the problems that the existing integrated landslide susceptibility evaluation method fails to effectively capture the dynamic evolution characteristics of geological processes, has limited applicability in rapidly changing environments, and has difficulty in high-precision feature extraction.

[0006] To achieve the above purposes, the technical scheme adopted by the present application is as follows: on the one hand, the present application provides an integrated landslide susceptibility evaluation method based on an XGB reinforcement model, comprising the following steps: S1, pre-process the collected evaluation factor data of the landslide disaster research area to obtain grid points, and extract the original data table of all evaluation factors corresponding to the landslide points by generating non-landslide points; S2, according to the original data table of all evaluation factors, the information value of the evaluation factor is calculated, the evaluation data set is constructed, and the evaluation data set is divided into a training set and a validation set according to a predetermined proportion; S3, construct a CatBoost-XGB integrated model, input the training set into the CatBoost-XGB integrated model for training, and verify it by using the validation set to obtain the optimal CatBoost-XGB integrated model; S4, import all grid point data in the region into the optimal CatBoost-XGB integrated model, perform landslide disaster susceptibility analysis, and output a landslide disaster susceptibility prediction map divided into a predetermined level.

[0007] The beneficial effects of the present application are: the present application fully utilizes the evaluation factor data through multi-level feature extraction and model integration, solves the problem of insufficient feature selection and weight distribution in traditional methods, significantly improves the precision and robustness of landslide susceptibility evaluation, and has high efficiency and reliability, which provides a new technical means for accurate evaluation and effective prevention and control of geological disasters; effectively capture the dynamic evolution characteristics of geological processes, and realize high applicability in rapidly changing environment, efficiently extract high-precision features.

[0008] Further, the S1 comprises the following steps: S101, collect the evaluation factor data of the landslide disaster research area; S102, unify the spatial resolution and coordinate system of the evaluation factor data, and obtain the same grid size layer by resampling to obtain the grid points; S103, according to the landslide points of the landslide disaster research area, randomly generate the same number of non-landslide points outside the preset diameter, and extract the original data table of all evaluation factors corresponding to the landslide points according to the preset extraction ratio.

[0009] The beneficial effects of the above further scheme are: the present application generates the same number of non-landslide samples outside the preset kilometer buffer zone of the landslide point through the spatial constraint sample equalization processing, considers the spatial distribution characteristics of the samples, effectively eliminates the spatial autocorrelation interference, and ensures the representativeness of the training data.

[0010] Further, the S2 comprises the following steps: S201, according to the original data table of all evaluation factors, remove the missing values and abnormal values in the original data table to obtain the processed original data table; S202, calculate information quantity values of each evaluation factor according to the processed original data table; S203, replace the original data of the evaluation factor with the information quantity value of the evaluation factor based on the processed original data table to obtain an evaluation data set; S204, set the evaluation factor data except the identity identification number and the attribute label as a preset label according to the evaluation data set, set the evaluation factor data containing the attribute label as a training target, and divide the evaluation data set into a training set and a verification set based on a preset proportion by using a random sampling method.

[0011] Further, the calculation expression of the information quantity value of each evaluation factor is as follows: ; Among them, represents the information quantity value, represents the landslide area of the current evaluation factor, represents the total landslide area of the study area, represents the area of the current evaluation factor, represents the total area of the study area.

[0012] The above further scheme has the beneficial effects that: the application quantifies the contribution degree of the evaluation factor by constructing an information quantity value dynamic weighting system, highlights the key evaluation factor, and makes the feature selection more scientific.

[0013] Further, the S3 comprises the following steps: S301, constructing a CatBoost-XGB integrated model comprising a primary model and a secondary model; S302, inputting the training set into the CatBoost-XGB integrated model to obtain a trained CatBoost-XGB integrated model; S303, verifying the performance of the trained CatBoost-XGB integrated model by using the verification set to obtain a verified CatBoost-XGB integrated model.

[0014] Further, the S301 comprises the following steps: S3011, importing a preset machine learning library and creating a chimeric model class; S3012, selecting a primary model CatBoost for feature selection and preliminary modeling and selecting a secondary model XGB for residual error optimization based on the fit method of the class; S3013, defining chimeric model parameters comprising primary model parameters and secondary model parameters, initializing the chimeric model class, and inputting the chimeric model parameters to obtain the CatBoost-XGB integrated model.

[0015] Furthermore, the prediction expression of the CatBoost-XGB integrated model is as follows: ; ; in, represents the predicted value of the XGB model, Indicates the k The prediction results of the decision tree, K represents the total number of decision trees, Represents the predicted value of the CatBoost model.

[0016] The beneficial effect of the above further scheme is as follows: the present invention constructs a cascaded CatBoost-XGB model architecture, first uses the CatBoost model to process category features and missing values, and then uses the XGBoost model to perform residual optimization, fully leveraging the advantages of the two models, improving the model's ability to mine data features, and improving the prediction accuracy of the evaluation model.

[0017] Furthermore, the S4 includes the following steps: S401, importing the original data of the evaluation factors in all grid point data in the region into the optimal CatBoost-XGB integrated model, predicting the landslide susceptibility of the study area, and outputting the susceptibility probability value of each grid cell in ascending order; S402. Based on the susceptibility probability values ​​of the grid cells arranged in ascending order, the natural breakpoint method is used to perform grade classification to complete the integrated landslide susceptibility assessment.

[0018] Furthermore, the specific steps for predicting the landslide susceptibility of the study area are: Based on the optimal CatBoost-XGB ensemble model, the CatBoost model was used for feature selection and preliminary modeling to obtain the importance and contribution of each evaluation factor in landslide susceptibility assessment. Based on the importance and contribution of each evaluation factor in landslide susceptibility assessment, in response to building a decision tree, the decision tree construction process is optimized using the XGB model to obtain an optimized decision tree; The optimized decision tree is used to output the landslide susceptibility probability value of each grid cell.

[0019] To achieve the above object, according to a second aspect of the present invention, there is provided an electronic device, characterized in that it includes a processor and a memory for storing executable instructions of the processor; The processor is configured to execute the above-mentioned integrated landslide susceptibility assessment method based on the XGB enhanced model.

[0020] To achieve the above object, according to a third aspect of the present application, a computer readable storage medium is provided, characterized in that the computer readable storage medium has stored thereon a plurality of classification programs for being invoked by a processor and executing the above-mentioned XGB reinforced model-based integrated landslide susceptibility assessment method.

[0021] The above-mentioned further scheme has the beneficial effect that the present application provides a high-precision and transferable intelligent assessment solution for regional landslide prevention and control by means of a non-landslide point generation strategy based on spatial constraints, a dynamic weighting system of information values, a cascade model design and a five-level susceptibility zoning map. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of the method of the present application.

[0023] Figure 2 An area under curve graph drawn in the present embodiment.

[0024] Figure 3 A histogram of performance indicators in the present embodiment.

[0025] Figure 4 A susceptibility evaluation map in the present embodiment. DETAILED DESCRIPTION

[0026] The specific embodiments of the present application are described below in order to facilitate the understanding of the present application by those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, any changes that are obvious within the spirit and scope of the present application as defined and determined by the appended claims are obvious, and all inventions utilizing the concept of the present application are within the scope of protection.

[0027] Before the present embodiment is described, the following terms are explained: XGB: XGBoost extreme gradient boosting; CatBoost: categorical boosting; fit method: a method for adjusting the internal parameters of a model according to the provided training data, so that the model can make predictions or decisions on unknown data; predict_proba method: a method for providing the predicted probability of each possible label; LightGBM: Light Gradient Boosting Machine; TWI: terrain wetness index; ROC curve: receiver operating characteristic curve; AUC: area under curve.

[0028] Embodiment 1 In this embodiment, an integrated landslide susceptibility assessment method combined with CatBoost-XGB model is proposed. This method fully utilizes multi-source data through multi-level feature extraction and model integration, effectively solves the problem of insufficient feature selection and weight allocation in traditional methods, and significantly improves the precision and robustness of landslide susceptibility assessment. In practical application, this method has successfully verified its efficiency and reliability, providing a new technical means for accurate assessment and effective prevention and control of geological disasters.

[0029] As shown in Figure 1 The present application provides an integrated landslide susceptibility assessment method based on XGB reinforced model, and the implementation method is as follows: S1, pre-processing the evaluation factor data of the collected landslide disaster research area to obtain grid points, and extracting the original data table of all evaluation factors corresponding to the landslide points by generating non-landslide points, the specific steps are as follows: S101, collecting evaluation factor data of landslide disaster research area; S102, unify the spatial resolution and coordinate system of the evaluation factor data, and obtain the same grid size layer by resampling to obtain the grid points; S103, according to the landslide point of the landslide disaster research area, randomly generate the same number of non-landslide points outside the preset diameter as the landslide points, and extract the original data table of all evaluation factors corresponding to the landslide points according to the preset extraction ratio.

[0030] In this embodiment, the present application selects Yuanling County area as the research area, integrates the data of local geological bureau, China Academy of Sciences Resource Environment Science and Data Center and Landslides_SL, and obtains 289 landslide points. In this embodiment, 16 evaluation factors are selected, including elevation, slope, slope direction, surface cutting coefficient, profile curvature, elevation variation coefficient, ground roughness, curvature, TWI, river distance, road distance, fault distance, normalized difference vegetation index (NDVI), land use type, lithology and rainfall; among them, the digital elevation model (DEM) data comes from geographic spatial data cloud, and the environmental factors obtained directly from the digital elevation model (DEM) include elevation, slope, slope direction, surface cutting coefficient, profile curvature, elevation variation coefficient, ground roughness, curvature and TWI; lithology and fault distance data come from China Geological Survey Bureau Geological Cloud; road and hydrological data, rainfall, normalized difference vegetation index (NDVI) come from geographic spatial data cloud; land use type data comes from the data set published by Wuhan University; The spatial resolution and coordinate system of the evaluation factor data were unified, and the layers of the same grid size were obtained by resampling to obtain grid points. Specifically, the coordinate system of the evaluation factor data was converted to WGS_1984_UTM_Zone_46N (World Geodetic System of 1984 - Transverse Mercator - Zone 46 of the Northern Hemisphere). The evaluation factor data was imported into the geographic information system Arcgis and the coordinate system of the evaluation factor data was unified based on the Data Management Tool - Projections and Transformations - Define Projection. The unified evaluation factor data was input into the geographic information system Arcgis again, and the multi-category data after the unified coordinate system was resampled through Data Management Tools—Raster—Raster Processing—Resample. The elevation raster was selected as the reference image, and the Input Raster was set to the raster image to be resampled. The Output Raster Dataset was set to "newraster1", that is, the raster image after resampling. The Output Cell Size was set to "Same as layer (same as layer) elevation raster", and Resample was set to "NEAREST" to obtain the converted raster data, that is, to obtain a layer of the same raster size and obtain the grid points. Based on the landslide points in the landslide hazard study area, the same number of non-landslide points are randomly generated outside the preset diameter of 1 km, and the original data table of all evaluation factors corresponding to the landslide points is extracted according to the preset extraction ratio: landslide points to non-landslide points 1:1.

[0031] S2. Based on the original data tables of all evaluation factors, the evaluation data set is constructed by calculating the information value of the evaluation factors, and the evaluation data set is divided into a training set and a validation set according to a preset ratio. The specific steps are as follows: S201, based on the original data table of all evaluation factors, removing missing values ​​and abnormal values ​​in the original data table to obtain a processed original data table; S202, calculating the information value of each evaluation factor based on the processed original data table; S203, based on the processed original data table, using the information value of the evaluation factor to replace the original data of the evaluation factor to obtain an evaluation data set; S204, according to the evaluation data set, the evaluation factor data except the identity identification number and the attribute label is set as a preset label, the evaluation factor data containing the attribute label is set as a training target, and the evaluation data set is divided into a training set and a verification set by using a random sampling method based on a preset proportion.

[0032] In this embodiment, according to the original data table of all evaluation factors, Excel is used for preprocessing to remove missing values and abnormal values in the original data table, and a processed original data table is obtained. According to the processed original data table, the information quantity value of each evaluation factor is calculated, and the expression is as follows: Among them, represents the information quantity value, represents the landslide area of the current evaluation factor, represents the total landslide area of the study area, represents the area of the current evaluation factor, represents the total area of the study area. The information quantity value of each evaluation factor is used to replace the original value of the evaluation factor, and a new data set, the evaluation data set, is generated. Based on the Python software, the data set is processed, the evaluation factor values except ID and attribute label Y are set as a preset label X, and the attribute data Y is a training target Y (1 for positive samples and 0 for negative samples); And a random sampling method is used to ensure that the distribution of the training set and the verification set is consistent, to avoid model performance deviation caused by uneven data distribution, and the evaluation data set is divided into a training set and a verification set according to a preset proportion of 7:3.

[0033] S3, a CatBoost-XGB integrated model is constructed, the training set is input into the CatBoost-XGB integrated model for training, and the verification set is used for verification to obtain an optimal CatBoost-XGB integrated model, and the specific steps are as follows: S301, a CatBoost-XGB integrated model containing a primary model and a secondary model is constructed, and the specific steps are as follows: S3011, import a preset machine learning library, and create an embedded model class; S3012, based on the fit method of the class, select the primary model CatBoost for feature selection and preliminary modeling, and select the secondary model XGB for residual error optimization; S3013, define the embedded model parameters containing the primary model parameters and the secondary model parameters, initialize the embedded model class, and input the embedded model parameters to obtain the CatBoost-XGB integrated model.​

[0034] In this embodiment, the libraries required for machine learning are imported, including CatBoostClassifier, XGBMClassifier, train_test_split, roc_curve, auc, accuracy_score, f1_score, precision_score, etc.; from sklearn.base module, import BaseEstimator and ClassifierMixin; from sklearn.utils.validation module, import functions for checking whether the input data meets the requirements (check_X_y, check_array, check_is_fitted); from sklearn.metrics module, import multiple index calculation functions for evaluating model performance, including area under the receiver operating characteristic curve calculation function (roc_curve, auc), accuracy calculation function (accuracy_score), F1 value calculation function (f1_score), precision calculation function (precision_score), recall calculation function (recall_score), classification report generation function (classification_report); from sklearn.ensemble module, import CatBoostClassifier class; import XGBoost library and alias it as XGB.

[0035] In this embodiment, a hybrid model class HybridModel is created, which inherits from BaseEstimator and ClassifierMixin, used to combine primary and secondary models; in the fit method of the class, use the primary model (choose CatBoost) for feature selection or preliminary modeling; in the predict_proba method of the class, use the secondary model (choose LightGBM set to XGB) for prediction; Define the initialization method (init), which initializes the primary model and its parameters, and the secondary model and its parameters respectively. If the corresponding model parameters are not provided, initialize them as empty dictionaries; In the fit method, receive input data X and target value y, and perform necessary checks, while storing the data in the class attributes for subsequent model training; In the predict_proba method, use the secondary model for prediction, output the predicted probability value; Define the parameters of the hybrid model, including the parameters of the primary model and the secondary model. Initialize the HybridModel class and pass in the parameters of the primary model and the secondary model in the parameters of the hybrid model to obtain the CatBoost-XGB ensemble model. The specific parameter settings are as follows: For CatBoost-XGBoost ensemble model: CatBoost is used as the primary model, and its corresponding parameters include: leaf regularization coefficient (l2_leaf_reg) is set to 3, learning rate (learning_rate) is set to 0.1, maximum depth (max_depth) is set to 9, number of estimators (n_estimators) is set to 200, verbose parameter is set to 0, random state (random_state) is set to 42; The secondary model uses XGBoost with the following parameters: gamma is set to 0.1, learning_rate is set to 0.2, maximum depth is set to 9, number of estimators is set to 100, and random_state is set to 42. Through the above steps, a CatBoost-XGBoost integrated model was successfully constructed. This model combines the advantages of CatBoost and XGBoost models and can effectively improve the accuracy and reliability of landslide susceptibility assessment.

[0036] In this embodiment, the prediction formula of XGBoost is the weighted sum of the prediction results of multiple trees, and its expression is as follows: ; in, represents the predicted value of the XGB model, Indicates the k The prediction results of the decision tree, K represents the total number of decision trees; The prediction formula of the CatBoost model is based on the integrated prediction of multiple decision trees, and its expression is as follows: ; in, Represents the predicted value of the CatBoost model; The objective function of the CatBoost model is based on the gradient boosting algorithm, which optimizes the model parameters by minimizing the loss function during the training process; the objective function of the XGBoost model is to improve the model performance by optimizing the loss function, while adding a regularization term to prevent overfitting; the above two models use the cross-entropy loss function in logistic regression, the expression of the loss function is: ; Wherein, represents the loss value, represents the number of samples in the training set, represents the true label of the i-th sample (1 represents landslide point, 0 represents non-landslide point), represents the probability of the i-th sample predicted by the model as a landslide point, the loss function is used to measure the difference between the predicted probability of the model and the true label; during the training process, the parameters of the model are adjusted by minimizing this loss function, so as to improve the prediction performance of the model.

[0037] S302, input the training set into the CatBoost-XGB integrated model to obtain the trained CatBoost-XGB integrated model; S303, verify the performance of the trained CatBoost-XGB integrated model by using the validation set, and obtain the validated CatBoost-XGB integrated model.

[0038] In this embodiment, the CatBoost-XGB integrated model is trained using the training set data, and during the training process, the primary model CatBoost is used for feature selection or preliminary modeling, and the secondary model XGB is used for residual optimization and final prediction; The performance of the trained CatBoost-XGB integrated model is verified by using the validation set, and the prediction performance of the model is evaluated by calculating AUC, accuracy (ACC), F1, precision (Precision), etc. to obtain the validated CatBoost-XGB integrated model, and the CatBoost-XGB integrated model after training and verification is saved to a file for subsequent use.

[0039] In this embodiment, the model accuracy is verified by using the validation set data, and the ROC curve is drawn to obtain the area AUC value of the curve as shown in Figure 2 The results show that the AUC value of the CatBoost-XGB chimeric model reaches 0.819, which is significantly better than the 0.664 of the single CatBoost model and the 0.749 of the single XGB model. At the same time, the performance of AUC, ACC, Precision and F1 is tested, and the results are as follows Figure 3As shown, the results show that the CatBoost-XGB chimeric model performs best in each performance indicator.

[0040] S4, import all grid point data in the region into the optimal CatBoost-XGB integrated model, and perform landslide disaster susceptibility analysis, and output a landslide disaster susceptibility prediction map divided into preset levels, the specific steps are as follows: S401, import the original data of the evaluation factors in all grid point data in the region into the optimal CatBoost-XGB integrated model, and perform landslide susceptibility prediction on the study area, and output the susceptibility probability value of each grid unit arranged in ascending order; S402, according to the susceptibility probability value of each grid unit arranged in ascending order, use the natural breakpoint method to divide the levels, and complete the integrated landslide susceptibility assessment.

[0041] In this embodiment, the information value of all evaluation factors of all grid points in the region is brought into the trained model, and the landslide susceptibility model of the optimal CatBoost-XGB integrated model is used to predict the landslide susceptibility of the study area. Based on the optimal CatBoost-XGB integrated model, the CatBoost model is used for feature selection and preliminary modeling to obtain the importance and contribution of each evaluation factor in landslide susceptibility assessment; Based on the importance and contribution of each evaluation factor in landslide susceptibility assessment, respond to the construction of decision tree, and use XGB model to optimize the decision tree construction process to obtain the optimized decision tree; Using the optimized decision tree, output the susceptibility probability value of each grid unit, and arrange it in ascending order; The natural breakpoint method is used to divide the susceptibility probability value into five levels: extremely low, low, medium, high and extremely high susceptible area. The extremely low, low, medium, high and extremely high susceptible area levels are self-divided and judged according to the actual situation.

[0042] In this embodiment, in general, the landslide susceptibility evaluation model in this embodiment has excellent performance; in the landslide geological disasters that have occurred, the landslide frequency ratio reaches 5.08; in addition, the AUC value of the CatBoost-XGB chimeric model is 0.819, indicating that its prediction effect is excellent, and the susceptibility evaluation map as shown is obtained. Figure 4

[0043] Embodiment 2 The embodiment provides an electronic device, including a processor and a memory for storing executable instructions of the processor; The processor is configured to implement the integrated landslide susceptibility assessment method based on the XGB reinforced model as described in Embodiment 1 by executing the executable instructions.

[0044] ​Example 3 This example proposes a computer-readable storage medium having stored thereon a plurality of classification programs for being invoked by a processor and executing the XGB reinforced model-based integrated landslide susceptibility assessment method as described in Example 1.

Claims

1. An integrated landslide susceptibility assessment method based on the XGB enhanced model, characterized by: The following steps are involved: S1. Preprocess the evaluation factor data collected in the landslide hazard study area to obtain grid points, and extract the original data table of all evaluation factors corresponding to the landslide points by generating non-landslide points; S2. Based on the original data tables of all evaluation factors, an evaluation data set is constructed by calculating the information value of the evaluation factors, and the evaluation data set is divided into a training set and a validation set according to a preset ratio; S3. Build a CatBoost-XGB ensemble model, input the training set into the CatBoost-XGB ensemble model for training, and use the validation set for validation to obtain the optimal CatBoost-XGB ensemble model. S4. Import all grid point data in the area into the optimal CatBoost-XGB integrated model to conduct landslide susceptibility analysis and output a landslide susceptibility prediction map divided into preset levels.

2. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 1 is characterized in that: The S2 comprises the following steps: S201, based on the original data table of all evaluation factors, removing missing values ​​and abnormal values ​​in the original data table to obtain a processed original data table; S202, calculating the information value of each evaluation factor based on the processed original data table; S203, based on the processed original data table, using the information value of the evaluation factor to replace the original data of the evaluation factor to obtain an evaluation data set; S204. According to the evaluation data set, the evaluation factor data other than the identity identification number and the attribute label are set as preset labels, and the evaluation factor data including the attribute label is set as the training target, and based on the preset ratio, the evaluation data set is divided into a training set and a validation set using a random sampling method.

3. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 2 is characterized in that: The calculation expression of the information value of each evaluation factor is as follows: in, Indicates the amount of information, represents the landslide area of ​​the current evaluation factor, represents the total landslide area in the study area, represents the area of ​​the current evaluation factor, Represents the total area of ​​the study area.

4. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 1 is characterized in that: The S3 includes the following steps: S301, constructing a CatBoost-XGB ensemble model including a primary model and a secondary model; S302, inputting the training set into the CatBoost-XGB ensemble model to obtain a trained CatBoost-XGB ensemble model; S303. Use the validation set to verify the performance of the trained CatBoost-XGB ensemble model to obtain a verified CatBoost-XGB ensemble model.

5. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 4 is characterized in that: The S301 includes the following steps: S3011. Import the preset machine learning library and create a chimeric model class; S3012, based on the class fit method, select the primary model CatBoost for feature selection and preliminary modeling, select the secondary model XGB for residual optimization, and use the chimeric model class to combine the primary model CatBoost and the secondary model XGB; S3013. Define chimeric model parameters including primary model parameters and secondary model parameters, initialize the chimeric model class, and input the chimeric model parameters to obtain a CatBoost-XGB integrated model.

6. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 5 is characterized in that: The prediction expression of the CatBoost-XGB ensemble model is as follows: in, represents the predicted value of the XGB model, Indicates the k The prediction results of the decision tree, K represents the total number of decision trees, Represents the predicted value of the CatBoost model.

7. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 1 is characterized in that: The S4 comprises the following steps: S401, importing the original data of the evaluation factors in all grid point data in the region into the optimal CatBoost-XGB integrated model, predicting the landslide susceptibility of the study area, and outputting the susceptibility probability value of each grid cell in ascending order; S402. Based on the susceptibility probability values ​​of the grid cells arranged in ascending order, the natural breakpoint method is used to perform grade classification to complete the integrated landslide susceptibility assessment.

8. The integrated landslide susceptibility assessment method based on the XGB enhanced model according to claim 7 is characterized in that: The specific steps for predicting landslide susceptibility in the study area are as follows: Based on the optimal CatBoost-XGB ensemble model, the CatBoost model was used for feature selection and preliminary modeling to obtain the importance and contribution of each evaluation factor in landslide susceptibility assessment. Based on the importance and contribution of each evaluation factor in landslide susceptibility assessment, in response to building a decision tree, the decision tree construction process is optimized using the XGB model to obtain an optimized decision tree; The optimized decision tree is used to output the landslide susceptibility probability value of each grid cell.

9. An electronic device, characterized in that: comprising a processor and a memory for storing executable instructions for the processor; The processor is configured to implement the integrated landslide susceptibility assessment method based on the XGB enhanced model according to any one of claims 1 to 8 by executing executable instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of classification programs, which are used to be called by a processor and execute the integrated landslide susceptibility assessment method based on the XGB enhancement model according to any one of claims 1 to 8.

Images