Geological disaster susceptibility cartographic model generation method and device

Abstract

The invention discloses a geological disaster susceptibility cartographic model generation method and device, and is used for solving the technical problem that a model cannot accurately reflect geological disaster susceptibility in practical application due to an existing training scheme of the model reflecting geological disaster susceptibility. The method comprises the steps of obtaining a geological disaster space-time training data set of a research area; adopting a preset data processing method to generate training data subsets corresponding to a plurality of grids in the research area according to the geological disaster space-time training data set; performing safety coefficient calculation according to the training data subset by adopting an infinite slope equation, and determining a training safety coefficient corresponding to each grid; and adopting a preset loss function to perform model training on the initial geological disaster susceptibility cartographic model according to the training safety coefficient and the training data subset corresponding to each grid, and determining a trained target geological disaster susceptibility cartographic model.

Description

Technical Field

[0001] The present invention relates to the technical field of address disaster prediction, and particularly to a method and device for generating a geological disaster susceptibility mapping model. Background Art

[0002] The susceptibility of geological disasters can be defined as the probability of geological disasters occurring at a specific location and time. Susceptibility is usually obtained through data-driven methods or physically based methods. Data-driven models provide effective statistical and machine learning (ML, Machine Learning) methods to establish a functional relationship between inputs and outputs, can directly infer the probability of a random process, and condition it with a set of predictor variables. They directly estimate the probability of geological disasters occurring given environmental conditions (including topography, geology, and other influencing factors).

[0003] However, the main problem with data-driven methods is that they often cannot accurately reflect the geological conditions of actual slope instability, and the models themselves do not respect the physical mechanisms behind the geological disaster process. For example, commonly used machine learning models, such as deep neural networks and tree-based models, are flexible and can capture complex patterns in large datasets.

[0004] Most of the existing training schemes for models reflecting geological disaster susceptibility focus on accuracy metrics of data-driven methods, such as precision, recall, and F1 score. However, due to the limitations of training data, these models may not fully reflect the underlying physical processes of geological disasters, thus neglecting the integration of data knowledge and physical mechanisms in the field of geological disaster susceptibility. And due to less attention to the interpretability of the models, the models cannot accurately reflect the geological disaster susceptibility in actual applications. Summary of the Invention

[0005] The present invention provides a method and device for generating a geological disaster susceptibility mapping model, which is used to solve the technical problem that the existing training schemes for models reflecting geological disaster susceptibility cannot accurately reflect the geological disaster susceptibility in actual applications.

[0006] A method for generating a geological disaster susceptibility mapping model provided in the first aspect of the present invention includes:

[0007] Obtain a spatio-temporal training dataset of geological disasters in the study area;

[0008] Use a preset data processing method to generate a training data subset corresponding to multiple grids in the study area according to the spatio-temporal training dataset of geological disasters;

[0009] Use the infinite slope equation to calculate the safety factor according to the training data subset, and determine the training safety factor corresponding to each grid;

[0010] The initial geological disaster susceptibility mapping model is trained by using a preset loss function according to the training safety factor corresponding to each grid and the training data subset, and the trained target geological disaster susceptibility mapping model is determined.

[0011] Optionally, the preset data processing method includes t-distributed stochastic neighbor embedding method and spatial cross-validation method; the step of using the preset data processing method to generate the training data subsets corresponding to multiple grids in the study area according to the spatio-temporal training data set of geological disasters includes:

[0012] The spatio-temporal training data set of geological disasters is screened by using the t-distributed stochastic neighbor embedding method to determine the spatio-temporal similar data set of geological disasters;

[0013] Based on the spatial cross-validation method, the spatio-temporal similar data set of geological disasters is divided to generate the training data subsets corresponding to multiple grids in the study area.

[0014] Optionally, the training data subset includes multiple surface static factors; the preset loss function includes a physics-based loss function and a binary cross-entropy loss function; the step of using the preset loss function to train the initial geological disaster susceptibility mapping model according to the training safety factor corresponding to each grid and the training data subset to determine the trained target geological disaster susceptibility mapping model includes:

[0015] The multiple surface static factors corresponding to each grid are input into the initial geological disaster susceptibility mapping model, and the initial geological disaster susceptibility values corresponding to each grid are output;

[0016] Based on the training safety factor corresponding to each grid, the grids are sorted in descending order to determine the grid sequence;

[0017] The geological disaster susceptibility differences between the initial geological disaster susceptibility values of adjacent grids in the grid sequence are calculated in sequence;

[0018] The first loss value is calculated by using the physics-based loss function according to the multiple geological disaster susceptibility differences;

[0019] The second loss value is calculated by using the binary cross-entropy loss function according to the multiple initial geological disaster susceptibility values;

[0020] The target loss value is determined according to the first loss value and the second loss value;

[0021] Based on the target loss value, the model parameters of the initial geological disaster susceptibility mapping model are updated to determine the intermediate geological disaster susceptibility mapping model, and the model update times are statistically counted in real time;

[0022] Determine whether the number of model updates reaches a preset training times threshold;

[0023] If it reaches, use the intermediate geological hazard susceptibility mapping model as the trained target geological hazard susceptibility mapping model.

[0024] Optionally, the training data subset further includes the water pressure head, soil depth, soil cohesion, and soil unit weight in the soil layer; the infinite slope equation is specifically:

[0025] ;

[0026] Where is the training safety factor at time t; is the soil cohesion, with the unit of Pa; is the soil unit weight, with the unit of N / m 2 , which is calculated from the soil bulk density and the gravitational acceleration ; ; is the soil depth, with the unit of m; is the specific gravity of water, with the unit of N / m 2 ; is the water pressure head in the soil layer; is the slope angle; is the soil internal friction angle.

[0027] Optionally, the multiple surface static factors include topographic factors, hydrological factors, environmental factor factors, and geological factors; the processing process of the initial geological hazard susceptibility value is specifically:

[0028] ;

[0029] Where LSM is the initial geological hazard susceptibility value; SNN is the initial geological hazard susceptibility mapping model; is the geological factor; is the topographic factor; is the hydrological factor; is the environmental factor factor.

[0030] Optionally, after the step of using the preset loss function to perform model training on the initial geological hazard susceptibility mapping model according to the training safety factor corresponding to each grid and the training data subset, and determining the trained target geological hazard susceptibility mapping model, it includes:

[0031] When receiving the geological hazard spatio-temporal prediction data set of the research area, use the preset data processing method to generate a prediction data subset corresponding to multiple grids in the research area according to the geological hazard spatio-temporal prediction data set;

[0032] Using the target geological disaster susceptibility mapping model, predictions are made based on multiple surface static factors in the prediction data subsets corresponding to the respective grids, and the target geological disaster susceptibility values corresponding to the respective grids are output;

[0033] Based on the multiple target geological disaster susceptibility values, a target geological disaster susceptibility map is constructed.

[0034] A device for generating a geological disaster susceptibility mapping model provided in the second aspect of the present invention includes:

[0035] An acquisition module for acquiring a spatio-temporal training data set of geological disasters in a study area;

[0036] An adoption module for generating training data subsets corresponding to multiple grids in the study area according to the spatio-temporal training data set of geological disasters by using a preset data processing method;

[0037] A calculation module for calculating a safety factor according to the training data subsets by using an infinite slope equation to determine the training safety factor corresponding to each grid;

[0038] A training module for training an initial geological disaster susceptibility mapping model by using a preset loss function according to the training safety factor corresponding to each grid and the training data subsets to determine a trained target geological disaster susceptibility mapping model.

[0039] A computer device provided in the third aspect of the present invention includes a memory and a processor. When a computer program stored in the memory is executed by the processor, the processor is caused to execute the steps of the geological disaster susceptibility mapping model generation method as described in any one of the above.

[0040] A computer-readable storage medium provided in the fourth aspect of the present invention has a computer program stored thereon. When the computer program is executed, the steps of the geological disaster susceptibility mapping model generation method as described in any one of the above are implemented.

[0041] A computer program product provided in the fifth aspect of the present invention includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer is caused to execute the steps of the geological disaster susceptibility mapping model generation method as described in any one of the above.

[0042] It can be seen from the above technical solutions that the present invention has the following advantages:

[0043] The above technical solution of the present invention provides a method for generating a geological hazard susceptibility mapping model. First, a spatio-temporal training data set of geological hazards in the study area is obtained; then, a preset data processing method is used to generate training data subsets corresponding to multiple grids in the study area according to the spatio-temporal training data set of geological hazards; the infinite slope equation is used to calculate the safety factor according to the training data subsets to determine the training safety factor corresponding to each grid; finally, a preset loss function is used to train the initial geological hazard susceptibility mapping model according to the training safety factor corresponding to each grid and the training data subsets to determine the trained target geological hazard susceptibility mapping model; based on the above solution, after processing the obtained spatio-temporal training data set of geological hazards by using the preset data processing method, the process of training the initial geological hazard susceptibility mapping model according to the training data subsets in combination with the infinite slope equation and the preset loss function to determine the trained target geological hazard susceptibility mapping model integrates the results of the physics-based model (infinite slope equation) of the present invention into an interpretable machine learning model (geological hazard susceptibility mapping model), which can follow the physical laws of the slope instability process, thereby improving the accuracy of the model in reflecting the geological hazard susceptibility in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0045] Figure 1 It is a flowchart of the steps of a method for generating a geological hazard susceptibility mapping model provided in Embodiment 1 of the present invention;

[0046] Figure 2 It is a flowchart of the steps of the actual application of the trained target geological hazard susceptibility mapping model provided in Embodiment 2 of the present invention;

[0047] Figure 3 It is a schematic diagram of the geological hazard risk susceptibility map of a certain area provided in Embodiment 2 of the present invention;

[0048] Figure 4 It is a verification schematic diagram of the performance of the data-driven SNN, PHY physical method, PHYSNN and ordinary deep learning LSTM in the geological hazard susceptibility mapping in four typical areas of arid, semi-arid, humid and semi-humid provided in Embodiment 2 of the present invention;

[0049] Figure 5Schematic diagram of the importance of each factor for the interpretability of geological hazard susceptibility mapping in a certain area provided in the second embodiment of the present invention;

[0050] Figure 6 Structural block diagram of a geological hazard susceptibility mapping model generation device provided in the third embodiment of the present invention. Detailed implementation manners

[0051] The embodiments of the present invention provide a method and device for generating a geological hazard susceptibility mapping model, which are used to solve the technical problem that the training scheme of the existing model reflecting the geological hazard susceptibility causes the model to be unable to accurately reflect the geological hazard susceptibility in actual applications.

[0052] In order to make the objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] Please refer to Figure 1 , Figure 1 Flowchart of the steps of a method for generating a geological hazard susceptibility mapping model provided in the first embodiment of the present invention.

[0054] A method for generating a geological hazard susceptibility mapping model provided by the present invention includes:

[0055] Step 101, obtain a spatio-temporal training data set of geological hazards in the study area.

[0056] It should be noted that taking each slope unit as the reference spatial scale, the present invention will aggregate the input prediction variables and the geological hazard stable / unstable label 1 at this scale. In order to identify the slope units where geological hazards exist, the present invention spatially overlaps the geological hazard trace area (excluding the sedimentation or flow area) with the slope units. If the geological hazard trace falls within the slope unit, then select this slope unit as the geological hazard-containing unit; otherwise, do not select. The reasons for the present invention to select the set of prediction variables are briefly summarized as follows:

[0057] Eastwardness: Eastwardness represents the degree to which the hillslope faces east. Northwardness: Similar to eastwardness, northwardness provides the degree to which the hillslope faces north. Horizontal curvature: Horizontal curvature provides information on the overall horizontal "bending" of a given slope unit relative to a virtual horizontal tangent and is a proxy for the three-dimensional geometric shape of the slope surface. Vertical curvature: Vertical curvature provides information on the overall vertical "bending" of the slope surface relative to a vertical line. Slope: Slope provides information on the steepness of the slope unit. Precipitation and soil moisture content: Provide information on the relative wetness of the slope surface for the model. The present invention includes the total precipitation and soil moisture content of the slope unit within the three months prior to the main event. NDVI: The Normalized Difference Vegetation Index (NDVI), as a proxy for the strength of vegetation roots, provides information on the overall quality and distribution of vegetation in the study area. PGA (Peak Ground Acceleration). Sand, silt, and clay content: This information provides the average content of sand, silt, and clay in the top 2 meters of soil layer of a given slope unit. Soil density: Similar to soil composition, soil density (or the oven-dried weight of dry soil) provides information on the composition of the failure surface material.

[0058] Furthermore, the model development process in the present invention includes data preparation, model construction, and performance evaluation for generating a geological hazard susceptibility map of the study area.

[0059] Step 102: Use a preset data processing method to generate a training data subset corresponding to multiple grids in the study area according to the geological hazard spatio-temporal training data set.

[0060] The preset data processing method includes the t-distributed stochastic neighbor embedding method and the spatial cross-validation method; the t-distributed stochastic neighbor embedding method is the t-Distributed Stochastic Neighbor Embedding (t-SNE) method; the spatial cross-validation method is the spatial cross-validation (Spatial CV).

[0061] It should be noted that to construct a comprehensive geological hazard spatio-temporal data set, historical typhoons, heavy rains, and geological hazards were integrated. Four types of geological hazard types, namely collapse, geological hazard, debris flow, and land subsidence, were considered. The present invention wants to emphasize that land subsidence is a subset of ground settlement, which reflects the settlement trend of the whole or local area. In terms of the sampling strategy, point sampling is adopted, and each point represents a grid with a spatial resolution of 30m, which is also the spatial resolution of the subsequent geological hazard susceptibility map. To ensure the model's ability to identify non-disasters, the present invention randomly selects the same number of non-disaster samples according to the proximity principle of each geological hazard during the corresponding period of the disaster occurrence.

[0062] Specifically, step 102 may include the following sub-steps S21-S22:

[0063] Step S21: Use the t-distributed Stochastic Neighbor Embedding method to screen the spatio-temporal training dataset of geological disasters and determine the spatio-temporal similar dataset of geological disasters;

[0064] Step S22: Based on the spatial cross-validation method, divide the spatio-temporal similar dataset of geological disasters to generate training data subsets corresponding to multiple grids in the study area.

[0065] It should be noted that the present invention uses the t-distributed Stochastic Neighbor Embedding (t-SNE) method to examine the distribution of the input dataset. t-SNE is a dimensionality reduction technique used to create a low-dimensional representation of high-dimensional data. The t-SNE map visually displays the structure and relationships of the data in two-dimensional or three-dimensional space by grouping similar data points closely together and separating dissimilar data points. This visualization method helps to reveal patterns, clusters, and potential outliers in the data and is a valuable tool. By using the t-distributed Stochastic Neighbor Embedding method to divide the spatio-temporal training dataset of geological disasters, a spatio-temporal similar dataset and a spatio-temporal dissimilar dataset of geological disasters are obtained. The spatio-temporal dissimilar dataset of geological disasters is excluded, and the spatio-temporal similar dataset of geological disasters is retained.

[0066] Furthermore, cross-validation (CV) is a commonly used model validation technique for evaluating the performance of machine learning models, especially effective in cases where data is limited. In this process, the dataset is divided into k subsets (i.e., folds). Each subset serves as the validation set, while the model is trained on the remaining k - 1 subsets. This process is repeated k times, with a different subset used as the validation fold each time. The final model performance is averaged over the performance metrics of all k iterations to ensure a robust assessment of the model performance and its stability on unseen data. However, general CV methods based on random sampling, such as random CV, usually assume that data samples are independent and identically distributed. This assumption may lead to overly optimistic performance estimates when applied to data with inherent temporal, spatial, hierarchical, or phylogenetic structures, which are common in fields such as ecology and geology / earth sciences.

[0067] To address this issue, it is better to divide the data into blocks that reflect its inherent structure (spatial autocorrelation in geological hazard data). This approach helps ensure the independence of the training and validation datasets and more accurately represents the complexity of the data. As an alternative, spatial cross-validation (spatial CV) is a method for evaluating the performance of prediction models in geospatial applications, including geological hazard detection and prediction. By ensuring the spatial independence of the training and validation sets, spatial CV can provide a more realistic assessment of model performance. In spatial CV, the dataset is divided into multiple non-overlapping spatial subsets (folds); for each fold, the model is trained on the remaining folds and tested on the target fold. Specifically, the spatio-temporal similarity dataset of geological hazards is divided by the spatial cross-validation method to obtain the training data subset corresponding to each grid in the study area. The extent of the ecological region is used as the dataset division strategy for spatial CV in this study, aiming to evaluate the generalization ability of the model in heterogeneous environments. In addition, the entire spatial CV process is repeated five times with different random seeds to accommodate and evaluate the uncertainty of the data and the model.

[0068] Step 103: Calculate the safety factor according to the training data subset using the infinite slope equation to determine the training safety factor corresponding to each grid.

[0069] The training data subset includes the water pressure head, soil depth, soil cohesion, and soil unit weight within the soil layer.

[0070] It should be noted that slope failures caused by rainfall are considered to be triggered by changes in pore pressure caused by rainfall. Therefore, stability analysis is usually combined with infiltration models to characterize the hydrological response of unsaturated slopes. The infiltration process is described by the Richards equation, which represents the flow of water in unsaturated soil. Assuming that the geometry and boundary conditions of the slope can be simplified to an infinite plane, the present invention selects the infinite slope equation to evaluate the stability of the slope body because it is one of the most widely used slope failure prediction models. This model is based on the assumption that the thickness of the sliding body (soil) is much smaller than the length of the slope, and this assumption is usually applicable to shallow geological hazards (more than 2 m deep). The safety factor FS is calculated as the ratio of the shear strength of the soil mass to the stress applied to the soil layer.

[0071] Furthermore, the infinite slope equation is specifically:

[0072] ;

[0073] where is the training safety factor at time t; is the soil cohesion, with the unit of Pa; is the soil unit weight, with the unit of N / m 2 , which is the bulk density of the soil mass and the acceleration of gravity Calculate ; is the soil depth, with the unit of m; is the specific gravity of water, with the unit of N / m 2 ; is the water pressure head in the soil layer; is the slope angle; is the soil internal friction angle.

[0074] It is worth mentioning that it is usually assumed that FS = 1 is the failure threshold. When FS < 1, that is, when the applied shear stress exceeds the shear strength of the soil mass, geological disasters occur. When FS ≥ 1, it indicates stability.

[0075] Step 104: Use the preset loss function to train the initial geological disaster susceptibility mapping model according to the training safety factor and training data subset corresponding to each grid, and determine the trained target geological disaster susceptibility mapping model.

[0076] The training data subset includes multiple surface static factors; the multiple surface static factors include topographic factors, hydrological factors, environmental factor factors, and geological factors.

[0077] The preset loss function includes a physics-based loss function and a binary cross-entropy loss function.

[0078] It should be noted that the present invention uses an interpretable additive artificial neural network model (Superposable Neural Network, SNN) as the geological disaster susceptibility mapping model, takes the surface static factors of geological disasters (such as elevation, slope, geology, fault distance, lithology, soil, vegetation cover, road network, water system, and forest loss, etc.) as independent variables, and geological disaster samples / non-geological disaster samples as dependent variables. The present invention uses the geological disaster observation points collected from 2000 to 2016 as the training set, and uses the observation points from 2017 to 2020 for verification. Using the trained target geological disaster susceptibility mapping model, finally draw an accurate static susceptibility map of geological disasters. This framework combines the model extraction method and the feature analysis-based method to ensure the comprehensiveness of the model, and optimizes the model by pruning redundant or suboptimal features and the interdependence between features. Each artificial neural network focuses on studying a specific feature, and finally optimizes the prediction effect of the entire model.

[0079] Furthermore, different from explainable artificial intelligence (XAI) methods such as Shapley Additive Explanation (SHAP) that only explain the local behavior of "black box" models, SNN has inherent global and local interpretability. In this framework, the goal of the model extraction method is to train an interpretable "student" model to mimic the behavior of the "teacher" model, while the feature analysis-based method is used to quantify the contribution of each input feature to the model result. This optimization framework not only has high interpretability but also performs excellently in terms of model accuracy, generalization, and complexity. In addition, based on techniques for dataset partitioning and result interpretation, the SNN model is particularly suitable for geological hazard susceptibility modeling applications with spatial dependence. The model further quantifies the relative importance of each feature for geological hazard susceptibility and analyzes the impact of the interaction between these features on the physical control mechanism of geological hazards, highlighting the underestimated important control factors. Overall, the geological hazard susceptibility mapping (LSM) based on the interpretable SNN method reveals the causal relationships between geological hazards and geological, hydrological, and meteorological factors at observation points and generalizes these relationships to unknown areas.

[0080] Specifically, step 104 may include the following sub-steps:

[0081] Step S41: Input multiple surface static factors corresponding to each grid into the initial geological hazard susceptibility mapping model, and output the initial geological hazard susceptibility values corresponding to each grid;

[0082] Step S42: Based on the training safety factor corresponding to each grid, perform a descending order sorting on each grid to determine the grid sequence;

[0083] Step S43: Calculate the geological hazard susceptibility difference of the initial geological hazard susceptibility values of adjacent grids in the grid sequence in turn;

[0084] Step S44: Use a physics-based loss function to calculate the first loss value according to multiple geological hazard susceptibility differences;

[0085] Step S45: Use a binary cross-entropy loss function to calculate the second loss value according to multiple initial geological hazard susceptibility values;

[0086] Step S46: Determine the target loss value according to the first loss value and the second loss value;

[0087] Step S47: Update the model parameters of the initial geological hazard susceptibility mapping model based on the target loss value to determine the intermediate geological hazard susceptibility mapping model, and count the model update times in real time;

[0088] Step S48: Determine whether the model update times reach the preset training times threshold;

[0089] Step S49: If the condition is met, use the intermediate geological hazard susceptibility mapping model as the trained target geological hazard susceptibility mapping model.

[0090] It should be noted that in the analysis of geological hazard stability, a monotonic relationship is usually accepted, that is, the relationship between geological hazard susceptibility and the safety factor (FS). FS measures the ratio of the resistance to the driving force along the potential failure surface. A higher FS value usually indicates lower geological hazard susceptibility. A basic physical relationship is established, and the model can learn and follow that when FS is high, the geological hazard susceptibility predicted by the model should be low; when FS is low, the geological hazard susceptibility predicted by the model should be high. However, a machine learning model trained only with data may not be able to reflect this physical relationship, resulting in a lack of scientific interpretability of the prediction results. If the model prediction violates this physical law, it means that the model has physical inconsistency and needs to be adjusted and optimized through a physically guided loss function. Therefore, it is ideal to guide the model towards physical consistency during the training process. The monotonic relationship between the model prediction value LSM and the FS values of any two samples can be expressed as:

[0091] ;

[0092] where, is the LSM value (initial geological hazard susceptibility value) corresponding to the first grid; is the training safety factor corresponding to the first grid.

[0093] Furthermore, if the samples in each training batch are sorted in descending order according to their FS values (i.e., FSi ≥ FSi+1), then the model prediction difference of any adjacent sample pair can be calculated, that is, based on the training safety factor corresponding to each grid, the grids are sorted in descending order to determine the grid sequence; successively calculate the geological hazard susceptibility difference of the initial geological hazard susceptibility values of adjacent grids in the grid sequence. The calculation process of the geological hazard susceptibility difference can be expressed as:

[0094] ;

[0095] where, is the geological hazard susceptibility difference; is the (i + 1)-th initial geological hazard susceptibility value; is the i-th initial geological hazard susceptibility value.

[0096] Furthermore, the negative value of can be regarded as a violation of the physical mechanism. A physics-based loss term L phy (i.e., the physics-based loss function) can be defined to measure the average value of these physical violations (i.e., physical inconsistency). The expression of the physics-based loss function is:

[0097] ;

[0098] wherein, is the first loss value; n is the total number of grids, i.e., the number of training samples; is the difference in geological disaster susceptibility.

[0099] Furthermore, (geological disaster susceptibility prediction) is regarded as a binary classification problem. This means that the goal of the model is to determine whether each sample belongs to the geological disaster susceptible area based on the input features, usually represented by 0 and 1: 1 indicates that the area is prone to geological disasters, and 0 indicates not prone.

[0100] The binary cross-entropy loss in the binary classification problem is the standard loss function used to measure the difference between the model prediction and the actual label. The formula of the binary cross-entropy loss function is as follows:

[0101] ;

[0102] wherein, is the second loss value; is the initial geological disaster susceptibility value of the i-th grid, i.e., the probability value predicted by the model, representing the probability that sample i belongs to class 1; is the true label (0 or 1) of the i-th grid, i.e., the true label (0 or 1) of the i-th sample.

[0103] It is worth mentioning that the model complexity loss (i.e., the regularization loss used to control the model complexity) may inadvertently weaken the emphasis on physical consistency, so this loss is not considered in the present invention.

[0104] Furthermore, traditional machine learning (ML) models (such as SNN) usually have difficulty accurately representing complex scientific relationships directly derived from data, especially when the training data is insufficient. In recent years, researchers have adopted physics-guided loss functions to address the challenges encountered by traditional ML models in various applications. This framework incorporates domain-specific scientific knowledge into the model and into the loss function as a regularization term, guiding the model towards physical consistency and enhanced performance.

[0105] The present invention introduces a physics-guided loss function (Lpg), which combines the traditional data loss (L data ) and the physics loss (Lphy). In this way, the model not only focuses on accuracy during optimization but also ensures that the results conform to physical consistency. Wherein, L datais the traditional supervised learning loss, which measures the difference between the model prediction and the true label; Lphy is the physical consistency loss, which measures whether the model prediction conforms to the known physical relationships. The calculation process of the target loss value is specifically as follows:

[0106] ;

[0107] Among them, is the target loss value; is the regularization loss for optimizing the model simplicity; L data represents the supervision error between the predicted value and the actual value, R is the regularization loss for optimizing the model simplicity, and L phy is the physics-based loss, which measures the consistency between the model prediction and the domain-specific scientific relationships. These three losses optimize the performance of the model in terms of accuracy, simplicity, and consistency respectively. The parameters and are hyperparameters that control the weights of R and L phy in the physics-guided loss function respectively. It should be noted that the first two are the standard losses of traditional machine learning models.

[0108] Furthermore, the processing process of the initial geological hazard susceptibility value is specifically as follows:

[0109] ;

[0110] Among them, LSM is the initial geological hazard susceptibility value; SNN is the initial geological hazard susceptibility mapping model; is the geological factor; is the terrain factor; is the hydrological factor; is the environmental factor.

[0111] Furthermore, if the number of model updates has not reached the preset training times threshold, the intermediate geological hazard susceptibility mapping model is used as the new initial geological hazard susceptibility mapping model, and then jump to execute step S41 until the number of model updates reaches the preset training times threshold. The intermediate geological hazard susceptibility mapping model determined when the number of model updates reaches the preset training times threshold is used as the trained target geological hazard susceptibility mapping model, and then it is verified on the test set.

[0112] It is worth mentioning that the evaluation performance of the model prediction results depends on the evaluation metrics of the test set. The likelihood of geological disasters is a typical binary classification problem, and the present invention uses 0.5 as the threshold to judge the predicted value. If the predicted value > 0.5, it is determined as a disaster (assigned a value of 1), otherwise it is determined as a non-disaster (assigned a value of 0). The accuracy (ACC), settlement amount (F), sensitivity (SST), specificity (SP), and F1 score (FS) are selected as the model evaluation metrics, and the defined formulas are as follows:

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] Among them, TP represents the positive number of the actual value and the predicted value. FP represents the number of actual negative values and positive predicted values. FN represents the number of actual positive values and negative predicted values. TN represents the number of cases where both the actual value and the predicted value are negative.

[0119] The values of ACC, P, SP, SST, and FS range from [0, 1]. When ACC, SP, SST, and FS approach 1, F approaches 0, indicating that the overall accuracy of the model is relatively high and the model has strong disaster prediction ability. In addition, the receiver operating characteristic (ROC) curve can evaluate the prediction ability of the model according to sensitivity and specificity at different probability thresholds. The area under the ROC curve is the AUC value, and its value range is 0.5 - 1. The AUC value representing an inaccurate model is less than 0.5. The closer the AUC value is to 1, the better the performance of the model. Generally speaking, if the AUC value is greater than 0.8, the performance of the model is good.

[0120] As a comparison of technical effects, it can be referenced in combination with the existing technology. Geological hazard susceptibility can be defined as the probability of a geological hazard occurring at a specific location and time. Susceptibility is usually obtained through data-driven methods or physically based methods. Data-driven models provide effective statistical and machine learning (ML) methods to establish a functional relationship between inputs and outputs, can directly infer the probability of a stochastic process, and condition it with a set of predictor variables. They directly estimate the probability of a geological hazard occurring given environmental conditions (including topography, geology, and other influencing factors). However, the main problem with data-driven methods is that they often cannot accurately reflect the geological conditions of actual slope instability, and the models themselves also fail to respect the physical mechanisms behind the geological hazard process. For example, commonly used machine learning models, such as deep neural networks and tree-based models, are flexible and can capture complex patterns in large datasets. However, this flexibility may lead to overfitting and poor generalization ability, especially when the training data is limited. In addition, these models may not reflect the underlying physical processes of geological hazards, such as mass movement dynamics, which poses more challenges in their practical applications.

[0121] Physically based methods have dominated current geotechnical research and practice, relying on physical laws derived from soil mechanics to determine slope stability conditions. In the field of geotechnical engineering, significant progress has been made in the research on analyzing slope stability using physically based methods. Physically based models can predict potential geological hazard areas through geological conditions without labeled data. The factor of safety (FS) has been widely used to quantify slope stability. However, due to the complexity of subsurface conditions of slopes, successfully determining the FS using physically based methods requires a large amount of field investigation and analysis, which limits the scale and efficiency of slope stability analysis and its application in risk management. In many cases, due to the difficulty in obtaining variables at the regional scale, physically based methods usually simplify variables at this scale, assuming that the geological properties within the study area are constant. Therefore, combining physically based models with data-driven methods is the key to improving the accuracy and reliability of geological hazard susceptibility models. In addition, the "black box" nature of the models makes it difficult for relevant experts to understand and trust their results. The lack of interpretability is a major obstacle in the practical application of geological hazard modeling, especially in high-risk fields such as geological hazards that involve public safety. In recent years, "explainable artificial intelligence" (XAI) models have been developed. These models can not only provide accuracy but also offer clear explanations to decision-makers. However, there is still a trade-off between accuracy and interpretability in current XAI technology.

[0122] To meet the demand for forecasting and early warning of geological disaster events under the current background of frequent extreme weather. Therefore, there is an urgent need for a risk early warning technology with generalization ability. Currently, the technology mainly focuses on the accuracy index of data-driven methods, ignoring the integration of data knowledge and physical mechanisms in the field of geological disaster susceptibility, and paying less attention to the interpretability of models, which is crucial for simulating complex processes such as geological disasters. Inaccurate or overfitted models may not accurately reflect the susceptibility of geological disasters in practical applications.

[0123] In summary, to meet the demand for forecasting and early warning of geological disaster events under the current background of frequent extreme weather. Therefore, there is an urgent need for a risk early warning technology with generalization ability. Currently, the technology mainly focuses on the accuracy index of data-driven methods, ignoring the integration of data knowledge and physical mechanisms in the field of geological disaster susceptibility, and paying less attention to the interpretability of models, which is crucial for simulating complex processes such as geological disasters. Inaccurate or overfitted models may not accurately reflect the susceptibility of geological disasters in practical applications. In view of this, the present invention proposes a method for generating a geological disaster susceptibility mapping model, develops a method for estimating the factor of safety (FS) value of slope instability by using a measured historical database and two domain knowledge-based models, and develops a spatially generalized data-driven and physically mechanism-coupled mechanism for disaster susceptibility technology, which can provide a fully interpretable superposable neural network (SNN) model while ensuring accuracy. The framework aims to achieve high interpretability, high accuracy, and high generalization by optimizing feature selection and removing redundant or suboptimal features for predicting slope stability conditions (i.e., failure 1 / stable 0).

[0124] Specifically, the present invention develops a method for estimating the factor of safety (FS) value of slope instability by using a measured historical database and two domain knowledge-based models, and develops a spatially generalized data-driven and physically mechanism-coupled mechanism for disaster susceptibility technology, which can provide a fully interpretable superposable neural network model (Superposable Neural Network, SNN) while ensuring accuracy. The framework aims to achieve high interpretability, high accuracy, and high generalization by optimizing feature selection and removing redundant or suboptimal features for predicting slope stability conditions (i.e., failure 1 / stable 0).

[0125] Integrate the results of a physics-based model (infinite slope model) into an interpretable machine learning model to improve the performance of geological disaster prediction. It can follow the physical laws of slope instability processes, which constitutes its primary advantage. Secondly, the present invention uses geotechnical properties as potential predictive variables, extracts these features from the data, and further provides an estimation of the physical constraints for the probability of geological disasters. This enables the present invention to conduct geological disaster modeling by combining physical constraints without relying on a large amount of geotechnical data. It has the ability to generalize in different geographical regions. At the same time, it optimizes the interpretability and accuracy of the model, overcoming the trade-off among mechanism, data-driven, and interpretability in traditional methods. Compared with existing interpretable methods, this method provides better interpretability without sacrificing accuracy by generating a Superposable Neural Network (SNN) model.

[0126] In summary, compared with the existing traditional methods for producing risk maps, the present invention has the following advantages:

[0127] Advantage 1: The present invention integrates the results of a physics-based model (such as the infinite slope equation) into an interpretable machine learning model, thereby improving the performance of geological disaster prediction. This method can follow the physical laws of slope instability processes and, by using geotechnical properties as potential predictive variables, extracts these features from the data to further provide an estimation of the physical constraints for the probability of geological disasters. Compared with traditional methods, this method can conduct geological disaster modeling by combining physical constraints without relying on a large amount of geotechnical data and has the ability to generalize across regions.

[0128] Advantage 2: By generating a Superposable Neural Network (SNN) model, this technology provides better interpretability while maintaining high accuracy. This overcomes the trade-off between accuracy and interpretability in existing Explainable Artificial Intelligence (XAI) methods and optimizes the interpretability and accuracy of the model. Compared with traditional methods for producing geological disaster susceptibility maps, this method not only improves the prediction accuracy but also makes the results easier to understand and interpret, thus enhancing the trust and usability of decision-makers.

[0129] In an embodiment of the present invention, the present invention provides a method for generating a geological disaster susceptibility mapping model. First, a spatio-temporal training data set of geological disasters in a study area is obtained; then, a preset data processing method is used to generate training data subsets corresponding to multiple grids in the study area according to the spatio-temporal training data set of geological disasters; the infinite slope equation is used to calculate the safety factor according to the training data subsets to determine the training safety factor corresponding to each grid; finally, a preset loss function is used to train the initial geological disaster susceptibility mapping model according to the training safety factor corresponding to each grid and the training data subsets to determine the trained target geological disaster susceptibility mapping model; based on the above solution, after processing the obtained spatio-temporal training data set of geological disasters by using the preset data processing method, the process of training the initial geological disaster susceptibility mapping model according to the training data subsets in combination with the infinite slope equation and the preset loss function to determine the trained target geological disaster susceptibility mapping model integrates the results of the physics-based model (infinite slope equation) of the present invention into an interpretable machine learning model (geological disaster susceptibility mapping model), which can follow the physical laws of the slope instability process, thereby improving the accuracy of the model in reflecting the geological disaster susceptibility in practical applications.

[0130] For better illustration, refer to Figure 2 , Figure 2 FIG. is a flowchart of the actual application steps of the trained target geological disaster susceptibility mapping model provided in the second embodiment of the present invention, which specifically includes:

[0131] Step 201, when receiving a spatio-temporal prediction data set of geological disasters in a study area, use a preset data processing method to generate prediction data subsets corresponding to multiple grids in the study area according to the spatio-temporal prediction data set of geological disasters.

[0132] Step 202, use the target geological disaster susceptibility mapping model to predict according to multiple surface static factors in the prediction data subsets corresponding to each grid, and output the target geological disaster susceptibility values corresponding to each grid.

[0133] Step 203, construct a target geological disaster susceptibility map according to multiple target geological disaster susceptibility values.

[0134] It should be noted that, please refer to Figures 3 - 5 , the present invention constructs a target geological disaster susceptibility map based on the target geological disaster susceptibility value corresponding to each grid. For example, for the geological disaster susceptibility mapping of a certain area as shown in Figure 3 , during the process of geological disaster susceptibility mapping, the performance of the data-driven SNN, PHY physical method, PHYSNN and ordinary deep learning LSTM of the present invention is verified for the geological disaster susceptibility mapping of this area in four typical areas of arid, semi-arid, humid and semi-humid, asFigure 4 As shown, finally, calculate the importance of each factor that affects the interpretability of geological hazard susceptibility mapping in this area, that is, the degree of contribution to the model prediction performance, such as Figure 5 shown.

[0135] In the embodiment of the present invention, the present invention integrates the results of a physics-based model (infinite slope equation) into an interpretable machine learning model, thereby improving the geological hazard prediction performance. Combining the above model training method can follow the physical laws of the slope instability process, and use geotechnical properties as potential prediction variables to extract these features from the data, and further provide a physically constrained estimate of the probability of geological disasters. Compared with traditional methods, this method can perform geological hazard modeling with physical constraints without relying on a large amount of geotechnical data and has the generalization ability across regions. At the same time, by generating an additive artificial neural network (SNN) model, this technology provides better interpretability while maintaining high accuracy. This overcomes the trade-off between accuracy and interpretability in existing explainable artificial intelligence (XAI) methods and optimizes the interpretability and accuracy of the model. Compared with traditional geological hazard susceptibility map production methods, this method not only improves the prediction accuracy, but also makes the results easier to understand and interpret, thus enhancing the trust and usage effect of decision-makers.

[0136] Please refer to Figure 6 , Figure 6 which is the structural block diagram of a geological hazard susceptibility mapping model generation device provided in Embodiment 3 of the present invention.

[0137] A geological hazard susceptibility mapping model generation device provided by the present invention includes:

[0138] An acquisition module 601, configured to acquire a spatio-temporal training data set of geological hazards in a research area;

[0139] An adoption module 602, configured to generate a training data subset corresponding to multiple grids in the research area according to the spatio-temporal training data set of geological hazards by using a preset data processing method;

[0140] A calculation module 603, configured to calculate a safety factor according to the training data subset by using an infinite slope equation, and determine a training safety factor corresponding to each grid;

[0141] A training module 604, configured to perform model training on an initial geological hazard susceptibility mapping model according to the training safety factor corresponding to each grid and the training data subset by using a preset loss function, and determine a trained target geological hazard susceptibility mapping model.

[0142] Furthermore, the preset data processing method includes the t-distributed stochastic neighbor embedding method and the spatial cross-validation method; the adoption module 602 is specifically configured to:

[0143] Use the t-distributed Stochastic Neighbor Embedding method to screen the spatio-temporal training dataset of geological disasters and determine the spatio-temporal similar dataset of geological disasters;

[0144] Based on the spatial cross-validation method, divide the spatio-temporal similar dataset of geological disasters to generate training data subsets corresponding to multiple grids in the study area.

[0145] Furthermore, the training data subset includes multiple surface static factors; the preset loss function includes a physics-based loss function and a binary cross-entropy loss function; the training module 604 is specifically used for:

[0146] Input the multiple surface static factors corresponding to each grid into the initial geological disaster susceptibility mapping model, and output the initial geological disaster susceptibility values corresponding to each grid;

[0147] Based on the training safety factor corresponding to each grid, sort each grid in descending order to determine the grid sequence;

[0148] Calculate the geological disaster susceptibility differences of the initial geological disaster susceptibility values of adjacent grids in the grid sequence in turn;

[0149] Use the physics-based loss function to calculate the first loss value according to multiple geological disaster susceptibility differences;

[0150] Use the binary cross-entropy loss function to calculate the second loss value according to multiple initial geological disaster susceptibility values;

[0151] Determine the target loss value according to the first loss value and the second loss value;

[0152] Update the model parameters of the initial geological disaster susceptibility mapping model based on the target loss value to determine the intermediate geological disaster susceptibility mapping model, and count the model update times in real time;

[0153] Judge whether the model update times reach the preset training times threshold;

[0154] If it reaches, use the intermediate geological disaster susceptibility mapping model as the trained target geological disaster susceptibility mapping model.

[0155] Furthermore, the training data subset also includes the water pressure head, soil depth, soil cohesion, and soil unit weight in the soil layer; the infinite slope equation is specifically:

[0156] ;

[0157] Among them, is the training safety factor at time t; is the soil cohesion, with the unit of Pa; is the unit weight of soil, with the unit of N / m 2 , which is calculated from the bulk density of the soil mass and the acceleration due to gravity ; ; is the soil depth, with the unit of m; is the specific gravity of water, with the unit of N / m 2 ; is the water pressure head within the soil layer; is the slope angle; is the angle of internal friction of the soil.

[0158] Furthermore, multiple surface static factors include topographic factors, hydrological factors, environmental factor factors, and geological factors; the processing process of the initial geological disaster susceptibility value is specifically as follows:

[0159] ;

[0160] Among them, LSM is the initial geological disaster susceptibility value; SNN is the initial geological disaster susceptibility mapping model; is the geological factor; is the topographic factor; is the hydrological factor; is the environmental factor factor.

[0161] In an alternative device embodiment, it further includes:

[0162] The first module is used to generate a predicted data subset corresponding to multiple grids in the study area according to the geological disaster spatio-temporal prediction data set by using a preset data processing method when receiving the geological disaster spatio-temporal prediction data set of the study area;

[0163] The second module is used to perform predictions according to multiple surface static factors in the predicted data subsets corresponding to each grid by using a target geological disaster susceptibility mapping model, and output the target geological disaster susceptibility value corresponding to each grid;

[0164] The third module is used to construct a target geological disaster susceptibility map according to multiple target geological disaster susceptibility values.

[0165] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described device and modules can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0166] The embodiment of the present invention also provides a computer device, including a memory and a processor, and a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the geological disaster susceptibility mapping model generation method in any of the above embodiments.

[0167] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program / instructions are stored. When the computer program / instructions are executed by a processor, the steps of the geological disaster susceptibility mapping model generation method in any of the above embodiments are implemented.

[0168] An embodiment of the present invention also provides a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, the steps of the geological disaster susceptibility mapping model generation method in any of the above embodiments are implemented.

[0169] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in an electrical, mechanical or other form.

[0170] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0171] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a geological disaster susceptibility mapping model, characterized in that: include: Obtain a spatiotemporal training dataset of geological hazards in the study area; Using a preset data processing method to generate training data subsets corresponding to a plurality of grids in the study area according to the spatiotemporal training data set of geological hazards; Using an infinite slope equation to calculate the safety factor according to the training data subset, and determining the training safety factor corresponding to each grid; The preset loss function is used to train the initial geological hazard susceptibility mapping model according to the training safety factor and the training data subset corresponding to each grid, and the trained target geological hazard susceptibility mapping model is determined.

2. The method for generating a geological disaster susceptibility mapping model according to claim 1, characterized in that: The preset data processing methods include t-distribution random neighbor embedding method and spatial cross-validation method; The method of using a preset data processing method to generate a training data subset corresponding to a plurality of grids in the study area according to the spatiotemporal training data set of geological hazards comprises: The t-distribution random neighbor embedding method is used to screen the geological disaster spatiotemporal training data set to determine a geological disaster spatiotemporal similar data set; The spatiotemporal similarity dataset of geological disasters is divided based on the spatial cross-validation method to generate training data subsets corresponding to multiple grids in the study area.

3. The method for generating a geological disaster susceptibility mapping model according to claim 1, characterized in that: The training data subset includes a plurality of surface static factors; the preset loss function includes a physics-based loss function and a binary cross entropy loss function; the preset loss function is used to perform model training on the initial geological hazard susceptibility mapping model according to the training safety factor corresponding to each grid and the training data subset, and a trained target geological hazard susceptibility mapping model is determined, including: Inputting a plurality of surface static factors corresponding to each of the grids into an initial geological disaster susceptibility mapping model, and outputting an initial geological disaster susceptibility value corresponding to each of the grids; Based on the training safety factor corresponding to each grid, the grids are sorted in descending order to determine a grid sequence; sequentially calculating the geological disaster susceptibility difference of the initial geological disaster susceptibility values ​​of adjacent grids in the grid sequence; Calculating a first loss value based on a plurality of geological disaster susceptibility differences using a physics-based loss function; Calculating a second loss value based on the plurality of initial geological disaster susceptibility values ​​using a binary cross entropy loss function; Determining a target loss value according to the first loss value and the second loss value; Based on the target loss value, the model parameters of the initial geological disaster susceptibility mapping model are updated, an intermediate geological disaster susceptibility mapping model is determined, and the number of model updates is counted in real time; Determine whether the model update times reaches a preset training times threshold; If achieved, the intermediate geological disaster susceptibility mapping model is used as the trained target geological disaster susceptibility mapping model.

4. The method for generating a geological disaster susceptibility mapping model according to claim 1, characterized in that: The training data subset also includes water pressure head in the soil layer, soil depth, soil cohesion and soil unit weight; the infinite slope equation is specifically: ； in, is the training safety factor at time t; is soil cohesion, unit is Pa; is the unit weight of soil, in N / m 2 , based on soil bulk density and the acceleration due to gravity calculate ; is the soil depth in m; is the specific gravity of water, in N / m 2 ; is the water pressure head in the soil layer; is the slope angle; is the soil internal friction angle.

5. The method for generating a geological disaster susceptibility mapping model according to claim 3, characterized in that: The plurality of surface static factors include topographic factors, hydrological factors, environmental factors and geological factors; the processing process of the initial geological disaster susceptibility value is specifically as follows: ； Among them, LSM is the initial geological disaster susceptibility value; SNN is the initial geological disaster susceptibility mapping model; For geological factors; For topographic factors; For hydrological factors; For environmental factors.

6. The method for generating a geological disaster susceptibility mapping model according to claim 1, characterized in that: After the step of using a preset loss function to train the initial geological hazard susceptibility mapping model according to the training safety factor and the training data subset corresponding to each grid to determine the trained target geological hazard susceptibility mapping model, the method includes: When receiving a spatiotemporal prediction data set of geological hazards in a study area, generating prediction data subsets corresponding to a plurality of grids in the study area according to the spatiotemporal prediction data set of geological hazards using the preset data processing method; Using the target geological disaster susceptibility mapping model to predict based on multiple surface static factors in the prediction data subset corresponding to each grid, output the target geological disaster susceptibility value corresponding to each grid; A target geological disaster susceptibility map is constructed based on the multiple target geological disaster susceptibility values.

7. A device for generating a geological disaster susceptibility mapping model, characterized in that: include: An acquisition module is used to obtain the spatiotemporal training dataset of geological hazards in the study area; An adopting module is used to generate a training data subset corresponding to a plurality of grids in the study area according to the spatiotemporal training data set of geological hazards by using a preset data processing method; A calculation module, used for calculating the safety factor according to the training data subset using an infinite slope equation, and determining the training safety factor corresponding to each grid; The training module is used to use a preset loss function to perform model training on the initial geological disaster susceptibility mapping model according to the training safety factor and training data subset corresponding to each grid, and determine the trained target geological disaster susceptibility mapping model.

8. A computer device, characterized in that: It includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for generating a geological hazard susceptibility mapping model as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for generating a geological hazard susceptibility mapping model as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the method for generating a geological hazard susceptibility mapping model as described in any one of claims 1-6.

Images