Karst surface collapse identification method and system based on GIS and machine learning

Abstract

The invention discloses a karst surface collapse identification method and system based on GIS and machine learning, and relates to the technical field of geological disaster monitoring. The method comprises the specific steps that a karst surface collapse geographic space database is constructed; collapse point locations and non-collapse point locations are selected from the karst surface collapse geographic space database to form an initial sample data set, n influence factors are screened, and a sample set is generated through a geographic detector; constructing a karst surface collapse susceptibility evaluation model based on a random forest algorithm through the initial sample data set and the sample set; dividing the karst surface collapse susceptible place into a plurality of grid units, and predicting the collapse probability of each grid by using the karst surface collapse susceptibility evaluation model; and importing the probability value of collapse prediction into a GIS platform, and assigning values to grid points according to spatial positions to obtain a karst surface collapse susceptibility index layer. According to the invention, through coupling of geographical detection and machine learning, efficient prediction of karst surface collapse is realized.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring technology, and in particular to a method and system for identifying karst ground collapse based on GIS and machine learning. Background Technology

[0002] In existing technologies, relying solely on the basic characteristics of ground subsidence and macroscopic hydrogeological analysis methods can only yield a general range of karst ground subsidence susceptibility, making precise target area delineation difficult. Furthermore, with technological advancements, machine learning algorithms based on big data have achieved significant results in exploring potential patterns and classifying data. However, the probability statistics and inference simulations based on various factors differ somewhat from the data statistical classification in karst ground subsidence susceptibility analysis. Therefore, for those skilled in the art, how to utilize machine learning technology to achieve accurate karst hazard identification is a pressing issue that needs to be addressed. Summary of the Invention

[0003] The purpose of this invention is to provide a karst ground collapse identification method and system based on GIS and machine learning to solve the problems mentioned in the background technology. It combines GIS spatial analysis to achieve quantitative and refined collapse risk classification, and through the coupling of geographic exploration and machine learning, it enables efficient prediction of karst ground collapse.

[0004] To achieve the above objectives, the present invention provides the following solution: On one hand, it provides a method for identifying karst ground collapse based on GIS and machine learning, the specific steps of which include the following:

[0005] Construct a geospatial database of karst ground collapse;

[0006] From the karst ground collapse geospatial database, collapse points and non-collapse points are selected to form an initial sample dataset. Then, n influencing factors are selected from the karst ground collapse geospatial database and a sample set is generated using a geographic detector.

[0007] A karst ground collapse susceptibility evaluation model was constructed using the initial sample dataset and the sample set based on the random forest algorithm.

[0008] The karst ground collapse study area is divided into several grid units, and the collapse probability of each grid unit is predicted using the karst ground collapse susceptibility evaluation model.

[0009] Import the predicted probability values ​​of the collapse into the GIS platform, and assign values ​​to the grid points according to their spatial location to obtain the karst ground collapse susceptibility index layer.

[0010] Preferably, the method further includes dividing the distribution of the collapse probability into multiple intervals using the natural discontinuity method, with each interval corresponding to a different susceptibility level, and the susceptibility level corresponding to a different degree of susceptibility to karst ground collapse.

[0011] Preferably, it also includes obtaining a karst land collapse susceptibility zoning map based on a random forest algorithm from a karst land collapse susceptibility index layer.

[0012] Preferably, the specific steps for constructing the karst ground collapse geospatial database are as follows:

[0013] Collect geospatial data on the main influencing factors in karst landslide-prone areas, including geological conditions, hydrological environment, topography, and human engineering data.

[0014] The geospatial data of the main controlling influencing factors were refined using the frequency ratio method to obtain the first influencing factor data;

[0015] The data of the first impact factor is filtered by a geographic detector and a recursive feature elimination method to obtain the data of the second impact factor.

[0016] The geospatial database of karst ground collapse is constructed based on the data of the second influencing factor.

[0017] Preferably, the step of constructing the karst ground collapse geospatial database based on the second influencing factor data is as follows:

[0018] The data of the second influencing factor were vectorized and then imported into the GIS platform;

[0019] A unified coordinate system is established in the GIS platform to perform data correction and registration;

[0020] The continuous data in the second influencing factor data are discretized using the natural discontinuity method, and the data are classified to obtain the karst ground collapse geospatial database.

[0021] Preferably, the calculation formula for refining the geospatial data of the main controlling influencing factors using the frequency ratio method is as follows:

[0022]

[0023] Where, n i Let s be the number of disasters occurring in the i-th category of a certain influencing factor, N be the total number of disasters, and s be the number of disasters. i Let S be the area or number of grid cells of the i-th category, and S be the total area or total number of grid cells of the study area.

[0024] Preferably, the RFC hyperparameters of the karst ground collapse susceptibility assessment model are optimized using a Bayesian optimization method; and the model quality of the karst ground collapse susceptibility assessment model is evaluated using a confusion matrix.

[0025] Preferably, the initial sample dataset is divided into a training set and a test set in a 3:1 ratio to train the random forest classifier; wherein, the training set is used for hyperparameter optimization, and the test set is used for performance evaluation of the trained karst ground collapse susceptibility assessment model.

[0026] On the other hand, a karst ground collapse identification system based on GIS and machine learning is provided, including a data acquisition module, a data preprocessing module, an evaluation model construction module, a prediction module, and a classification module; among which,

[0027] The data acquisition module is used to construct a geospatial database of karst ground collapse;

[0028] The data preprocessing module is used to randomly select collapse points and non-collapse points from the karst ground collapse geospatial database to form an initial sample dataset, and to select n influencing factors from the karst ground collapse geospatial database and generate a sample set using a geographic detector.

[0029] The evaluation model construction module is used to construct a karst ground collapse susceptibility evaluation model based on the random forest algorithm using the initial sample dataset and the sample set.

[0030] The prediction module is used to divide the karst land collapse prone area into several grid cells and use the karst land collapse susceptibility evaluation model to predict the collapse probability of each grid cell.

[0031] The classification module is used to import the probability value of the collapse prediction into the GIS platform and assign values ​​to the grid points according to the spatial location to obtain the karst ground collapse susceptibility index layer.

[0032] According to specific embodiments provided by the present invention, the following technical effects are disclosed: Compared with traditional methods, machine learning technology, with its nonlinear data mining capabilities and high-dimensional pattern recognition advantages, can overcome the limitations of human experience and reveal potential collapse-causing patterns through statistical correlation of multi-source data. This method ensures the model's generalization ability through hyperparameter optimization and cross-validation, and achieves quantitative and refined collapse risk classification by combining GIS spatial analysis, providing efficient technical support for karst disaster prevention and control. The coupling of geographic exploration and machine learning makes the prediction of karst ground collapse more efficient and reliable. Attached Figure Description

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 This is a flowchart of the method of the present invention;

[0035] Figure 2 This is a flowchart illustrating the construction process of the karst ground collapse geospatial database of the present invention.

[0036] Figure 3 Flowchart for multi-source data entry into the geospatial database of karst ground collapse according to the present invention;

[0037] Figure 4 This is a flowchart illustrating the construction process of the karst ground collapse susceptibility evaluation model of the present invention. Detailed Implementation

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] The purpose of this invention is to provide a method for identifying karst ground collapse based on GIS and machine learning, characterized in that, as Figure 1 As shown, the specific steps include the following:

[0040] S1. Construct a geospatial database of karst ground collapse;

[0041] S2. Select collapse points and non-collapse points from the karst ground collapse geospatial database to form an initial sample dataset. Select n influencing factors from the karst ground collapse geospatial database and generate a sample set using a geographic detector.

[0042] S3. Construct a karst ground collapse susceptibility evaluation model based on the random forest algorithm using the initial sample dataset and sample set;

[0043] S4. Divide the karst ground collapse study area into several grid units and use the karst ground collapse susceptibility evaluation model to predict the collapse probability of each grid unit.

[0044] S5. Import the predicted probability value of the collapse into the GIS platform, assign values ​​to the grid points according to the spatial location, and obtain the karst ground collapse susceptibility index layer.

[0045] The formation of karst ground subsidence requires three main elements: solution space, loose overburden, and inducing factors. The easily collapsible geological structure formed by the combination of solution space and overburden is relatively stable under natural conditions. Subsidence formation is often the result of a combination of multiple inducing mechanisms. The main controlling factors of ground subsidence susceptibility can be summarized into four categories: ① geological conditions; ② topography; ③ hydrological environment; ④ human engineering activities. Based on the formation mechanism and considering the principle of data availability, this invention quantitatively selects easily accessible influencing factors from numerous factors to construct a susceptibility evaluation model. Furthermore, a geospatial database of karst ground subsidence is constructed through multi-source spatial data integration and preprocessing. Further, such as... Figure 2 As shown, the specific steps for constructing the karst ground collapse geospatial database in step S1 are as follows:

[0046] S11. Collect geospatial data of the main influencing factors in karst landslide-prone areas, including geological condition data, hydrological environment data, topographic data, and human engineering data.

[0047] S12. The geospatial data of the main influencing factors are refined using the frequency ratio method to obtain the data of the first influencing factor.

[0048] Specifically, the frequency ratio method is used to reflect the correlation between the category of influencing factors and geological hazards. The calculation formula is as follows:

[0049]

[0050] Where, n i Let s be the number of disasters occurring in the i-th category of a certain influencing factor, N be the total number of disasters, and s be the number of disasters. i Let be the area or number of grid cells of the i-th category, and S be the total area or total number of grid cells of the study area. i / N represents the proportion of collapse in the i-th category of a certain influence factor, s i / S represents the area proportion of this category. A larger FR proportion indicates more ground subsidence in a relatively small area, reflecting that this category is favorable for subsidence development. A smaller FR proportion indicates a lower correlation between this category and subsidence development.

[0051] S13. The data of the first impact factor is filtered by a geographic detector and a recursive feature elimination method to obtain the data of the second impact factor.

[0052] In this embodiment, the "differentiation and factor detection" detector in the geographic detector is used to determine the relationship between the variable (X) and the predicted result (Y) of the detection target, that is, the relationship between the influencing factor and karst ground collapse. This detector can explore the spatial differentiation of Y and the extent to which the variable X explains the spatial differentiation performance of Y, and uses the q-value to quantitatively explain it. The calculation formula is as follows:

[0053]

[0054] Where h represents the stratification of the impact factor X, and stratification is a statistical concept that corresponds to a geographical class or sub-region, representing the category of the impact factor in this invention; N is the number of units in the entire region; σ is the variance of h; 2 Let q be the total variance of Y values ​​across the entire region, where the value of q ranges from [0, 1]. A value of q = 1 indicates that the influencing factor fully explains the spatial distribution of the dependent variable, while a value of q = 0 indicates that the influencing factor has no relation to the spatial distribution of the dependent variable. The larger the value of q, the stronger the explanatory power of the influencing factor, i.e., the greater its importance to the spatial distribution of the dependent variable.

[0055] S14. Construct a geospatial database of karst surface collapse based on the second influencing factor data, such as... Figure 3 As shown, the specific steps are as follows:

[0056] S141. After vectorizing the data of the second influencing factor (such as remote sensing data, topographic maps, geological maps and field survey data), import it into the GIS platform.

[0057] S142. Establish a unified coordinate system in the GIS platform and perform data correction and registration;

[0058] The import process of original maps involves data preprocessing such as vectorization and coordinate transformation, which can cause data position shifts. Therefore, it is necessary to correct each layer on the GIS platform to eliminate data position errors. By using a base map to register multi-source data layers, the position of the same point can be kept consistent across different layers.

[0059] S143. The continuous data in the second influencing factor data is discretized by using the natural discontinuity method to maximize the difference between each class and unify the data type. The data is then classified to obtain a geospatial database of karst ground collapse.

[0060] It should be noted that in this invention, the area and grid cell count in the geospatial database construction are processed using the ArcGIS platform. High-resolution DEM data is georeferenced through projection transformation, and mask extraction is used to delineate the boundaries of the working area. Some vector data is converted to raster data. Then, spatial analysis methods such as regional analysis, extraction analysis, and surface analysis are used to generate various influencing factor layers. However, due to the multi-source and variability of the data, the size and number of grid cells in each layer will differ. But when statistically analyzing the attributes of collapse points, the nearest neighbor calculation can be used to assign values ​​of each influencing factor to each collapse point. In addition, data types are divided into discrete and continuous types. For example, stratigraphic lithology is discrete data, while various data extracted from the DEM are continuous by default. Since most influencing factors are related to natural factors, and karst ground collapse is a geological hazard caused by both natural and human factors, the natural breakpoint method or equidistant division method is used to divide the continuous data into intervals, resulting in discrete data.

[0061] Furthermore, in step S3, a karst ground subsidence susceptibility assessment model based on GIS and the random forest algorithm is constructed. The random forest algorithm is used for classification. When creating training samples, the target predicted value can be represented by 0 and 1, that is, the ground subsidence attribute is input as 0 (representing no subsidence) and 1 (representing subsidence). Then, the random forest classifier (RFC) is used to build a ground subsidence prediction model, such as... Figure 4 As shown, the specific steps are as follows:

[0062] S31. Random sampling is used to generate a sample of non-collapse points with an equal number of points. The attribute values ​​of each factor in the karst ground collapse geospatial database are assigned to the sample points through spatial location to form an initial sample dataset. Based on the n influencing factors selected from the working area, a sample set is generated using a geographic detector.

[0063] Specifically, the interpretability of each major influencing factor in the geospatial heterogeneity of the karst surface collapse geospatial database in the work area is ranked. In the above formula, the q-value represents the explanatory power. Based on the ranking results according to the size of the q-value, influencing factors are successively eliminated in ascending order of q-value, thereby constructing sample datasets. If the work area has n influencing factors, the first sample dataset has n influencing factors. Then, the influencing factor with the smallest q-value is eliminated, and the remaining n-1 influencing factors are used to construct the second sample dataset, and so on, until only the influencing factor with the largest q-value remains to construct the nth sample dataset. The n influencing factors correspond to the n sample datasets.

[0064] S32. The initial sample dataset is divided into training and test sets in a 1:3 ratio for model training. The training set is used for hyperparameter optimization, and the test set is used to evaluate the performance of the model trained with the optimal hyperparameters. In this invention, the RFC hyperparameters of the karst ground collapse susceptibility assessment model are optimized using the Bayesian optimization method; the confusion matrix is ​​used to evaluate the model quality of the karst ground collapse susceptibility assessment model; this invention establishes a method for factor and parameter coupling optimization to improve the prediction accuracy of the ground collapse RFC prediction model.

[0065] Furthermore, the specific process of optimizing the RFC hyperparameters of the karst ground collapse susceptibility assessment model using the Bayesian optimization method is as follows:

[0066] The goal of machine learning model optimization is to find the global optimum or a better local solution. Optimizing hyperparameters aims to find the combination of hyperparameters that best performs the model on the validation dataset. This invention selects the Bayesian optimization method to optimize the hyperparameters of the RFC (Real-Time Function). The basic idea of ​​Bayesian optimization is as follows: given the objective function to be optimized (which can be a generalized function), only the input and output need to be specified; the internal structure and mathematical properties of the function do not need to be known. The posterior distribution of the objective function is updated by continuously adding sample points until the posterior distribution substantially approximates the true distribution.

[0067] Bayesian optimization is an approximation theory. The advantages of Bayesian hyperparameter tuning include fewer parameter iterations and less time consumption. It estimates the posterior distribution of the objective function based on data using Bayes' theorem, and the next input is based on the posterior distribution of the previous input, thus making the result closer to the true posterior distribution, rather than simply exhaustively searching. Its optimization method involves learning the shape of the objective function and finding the parameters that maximize the result towards the global maximum. However, Bayesian optimization is prone to exploring local optima, potentially missing better solutions in unsampled regions. Optimizing a probability function can effectively address this problem. It represents the probability that the objective function value will be much larger than the current optimum. Let X be the hyperparameter to be optimized, and the probability optimization function formula is as follows:

[0068]

[0069] Where POI is the optimization probability function; f(X) is the objective function value of X; f(X) +Let f(x) be the current optimal objective function value; ξ be the tradeoff coefficient; μ(x) be the mean of the objective function; and σ(x) be the variance of the objective function. ξ in the equation affects whether the next sampling point is sampled from the current optimal solution or from an unsampled region. Without the tradeoff coefficient, the objective function would tend to sample near the current optimal value rather than exploring unsampled regions. The ultimate goal of the probabilistic optimization function is to continuously sample new X. Since the calculation of f(X) is complex, to determine the optimal X, the maximum value of POI(X) can be calculated. Therefore, we consider using the X corresponding to the maximum optimal probability function value as the final optimization result.

[0070] Furthermore, the specific process of using the confusion matrix for model quality evaluation is as follows:

[0071] In this embodiment of the invention, the confusion matrix, commonly used in binary classification models in machine learning algorithms, is employed for model quality evaluation. Positive examples represent points that have collapsed, while negative examples represent points that have not collapsed. A true positive (TP) indicates a collapsed point is predicted as collapsed; a true negative (TN) indicates a point that has not collapsed is correctly predicted as not collapsed; a false positive (FP) indicates a point that has not collapsed is incorrectly predicted as collapsed; and a false negative (FN) indicates a point that has collapsed is incorrectly predicted as not collapsed. The confusion matrix table is shown in Table 1.

[0072]

[0073] Table 1

[0074] Based on the confusion matrix, the following two model evaluation metrics are established:

[0075]

[0076] Here, Accuracy represents model precision, which means the proportion of samples that are correctly predicted out of all samples, i.e., positive examples are predicted as positive examples and negative examples are predicted as negative examples out of the total predicted samples. Precision represents accuracy, which means the proportion of samples that are predicted as collapsed actually collapsed out of all samples.

[0077] Since machine learning models output predictions for classification problems in probabilistic form, to build a confusion matrix and calculate accuracy, the probabilities need to be converted into predicted categories. This requires setting a probability threshold: if the probability is higher than the threshold, the predicted category is A; if the probability is lower, the predicted category is B. The probability threshold setting significantly impacts model accuracy calculations; therefore, the Receiver Operating Characteristic (ROC) curve is introduced. The ROC curve is also frequently used to validate classification models, especially when there is sample imbalance, where the ROC curve is almost unaffected. Sample imbalance refers to the situation where, when testing the model on the test set, there may be significantly more negative examples than positive examples. The horizontal and vertical axes of the ROC curve represent the Free Response Rate (FPR) and Total Response Rate (TPR), respectively, and the calculation formula is as follows:

[0078]

[0079] The Free Predictive (FPR) curve measures how many negative examples were incorrectly predicted as positive, while the Total Predictive (TPR) curve measures how many positive examples were correctly predicted as positive. FPR focuses only on negative samples and is unrelated to positive samples, while the TPR curve focuses only on positive samples and is unrelated to negative samples. This reasonably explains why the ROC curve is not affected by sample imbalance. The ROC curve's shape is the only way to qualitatively judge the model's quality; the closer the curve is to the top left corner, the better the model's quality. Therefore, the Area Under the Curve (AUC) is introduced to quantitatively evaluate the model's performance; a higher AUC value indicates a more effective model. When AUC = 1, the binary classification model is perfect, while when AUC < 0.5, the model is meaningless because the AUC value for random prediction of a binary event is approximately 0.5.

[0080] Furthermore, the process involves dividing the work area into n raster cells (the raster cells are determined according to the scale requirements of the specific project for the work area's susceptibility assessment). The geospatial database's factor attributes are imported into Python, and a trained Random Forest Classifier (RFC) model is used to predict the collapse probability (0-1) of each raster. All predicted collapse probability values ​​are then imported into the GIS platform software, and values ​​are assigned to raster points based on their spatial location to obtain the karst ground collapse susceptibility index layer. To maximize the difference between different susceptibility levels, the collapse probability distribution is divided into four intervals using the natural discontinuity method, each interval corresponding to a susceptibility level. The susceptibility index is divided into four levels to correspond to different degrees of karst ground collapse susceptibility: high susceptibility, medium susceptibility, low susceptibility, and extremely low susceptibility. This allows for the generation of karst ground collapse susceptibility zoning maps based on the Random Forest algorithm.

[0081] Traditional methods based on basic characteristics of ground subsidence and macroscopic hydrogeological analysis can identify potential karst subsidence-prone areas, but their experience-driven qualitative judgments are difficult to apply to fine-grained zoning, especially under complex geological conditions, leading to blurred prediction boundaries. Compared to traditional methods, machine learning techniques, with their nonlinear data mining capabilities and high-dimensional pattern recognition advantages, can overcome the limitations of human experience and reveal potential subsidence-causing patterns through statistical correlation of multi-source data. This method ensures model generalization ability through hyperparameter optimization and cross-validation, and combines GIS spatial analysis to achieve quantitative and refined subsidence risk classification, providing efficient technical support for karst disaster prevention and control. The selection of indicator factors is reasonable, scientific, and operable; hyperparameter optimization and cross-validation improve the model's accuracy and universality; and the coupling of geographic exploration and machine learning makes the prediction of karst ground subsidence more efficient and reliable.

[0082] On the other hand, a karst ground collapse identification system based on GIS and machine learning is provided, including a data acquisition module, a data preprocessing module, an evaluation model construction module, a prediction module, and a classification module; among which,

[0083] The data acquisition module is used to construct a geospatial database of karst ground collapse.

[0084] The data preprocessing module is used to randomly select collapsed and non-collapsed locations from the karst ground collapse geospatial database to form an initial sample dataset, and to select n influencing factors from the karst ground collapse geospatial database and generate a sample set using a geographic detector.

[0085] The evaluation model building module is used to construct an evaluation model for the susceptibility of karst ground collapse based on the random forest algorithm using an initial sample dataset and a sample set.

[0086] The prediction module is used to divide the karst land collapse-prone area into several grid cells and use the karst land collapse susceptibility assessment model to predict the collapse probability of each grid cell.

[0087] The classification module is used to import the probability values ​​of the collapse prediction into the GIS platform and assign values ​​to the grid points according to their spatial location to obtain the karst ground collapse susceptibility index layer.

[0088] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying karst ground collapse based on GIS and machine learning, characterized in that, The specific steps include the following: Construct a geospatial database of karst ground collapse; From the karst ground collapse geospatial database, collapse points and non-collapse points are selected to form an initial sample dataset. Then, n influencing factors are selected from the karst ground collapse geospatial database and a sample set is generated using a geographic detector. A karst ground collapse susceptibility evaluation model was constructed using the initial sample dataset and the sample set based on the random forest algorithm. The karst ground collapse study area is divided into several grid units, and the collapse probability of each grid unit is predicted using the karst ground collapse susceptibility evaluation model. Import the predicted probability values ​​of the collapse into the GIS platform, and assign values ​​to the grid points according to their spatial location to obtain the karst ground collapse susceptibility index layer.

2. The method for identifying karst ground collapse based on GIS and machine learning according to claim 1, characterized in that, It also includes dividing the distribution of the collapse probability into multiple intervals using the natural discontinuity method, with each interval corresponding to a different susceptibility level, and the susceptibility level corresponding to a different degree of susceptibility to karst ground collapse.

3. The method for identifying karst ground collapse based on GIS and machine learning according to claim 1, characterized in that, It also includes a karst land collapse susceptibility zoning map based on the random forest algorithm, obtained from the karst land collapse susceptibility index layer.

4. The method for identifying karst ground collapse based on GIS and machine learning according to claim 1, characterized in that, The specific steps for constructing the karst ground collapse geospatial database are as follows: Collect geospatial data on the main influencing factors in karst landslide-prone areas, including geological conditions, hydrological environment, topography, and human engineering data. The geospatial data of the main controlling influencing factors were refined using the frequency ratio method to obtain the first influencing factor data; The data of the first impact factor is filtered by a geographic detector and a recursive feature elimination method to obtain the data of the second impact factor. The geospatial database of karst ground collapse is constructed based on the data of the second influencing factor.

5. The method for identifying karst ground collapse based on GIS and machine learning according to claim 4, characterized in that, The steps for constructing the karst ground collapse geospatial database based on the second influencing factor data are as follows: The data of the second influencing factor were vectorized and then imported into the GIS platform; A unified coordinate system is established in the GIS platform to perform data correction and registration; The continuous data in the second influencing factor data are discretized using the natural discontinuity method, and the data are classified to obtain the karst ground collapse geospatial database.

6. The method for identifying karst ground collapse based on GIS and machine learning according to claim 3, characterized in that, The calculation formula for refining the geospatial data of the main controlling influencing factors using the frequency ratio method is as follows: Where, n i Let s be the number of disasters occurring in the i-th category of a certain influencing factor, N be the total number of disasters, and s be the number of disasters. i Let S be the area or number of grid cells of the i-th category, and S be the total area or total number of grid cells of the study area.

7. The method for identifying karst ground collapse based on GIS and machine learning according to claim 1, characterized in that, The RFC hyperparameters of the karst ground collapse susceptibility assessment model were optimized using the Bayesian optimization method; the model quality of the karst ground collapse susceptibility assessment model was evaluated using the confusion matrix.

8. The method for identifying karst ground collapse based on GIS and machine learning according to claim 7, characterized in that, The initial sample dataset is divided into a training set and a test set in a 3:1 ratio to train the random forest classifier; the training set is used for hyperparameter optimization, and the test set is used to evaluate the performance of the trained karst ground collapse susceptibility assessment model.

9. A karst ground collapse identification system based on GIS and machine learning, characterized in that, It includes a data acquisition module, a data preprocessing module, an evaluation model construction module, a prediction module, and a classification module; among which, The data acquisition module is used to construct a geospatial database of karst ground collapse; The data preprocessing module is used to randomly select collapse points and non-collapse points from the karst ground collapse geospatial database to form an initial sample dataset, and to select n influencing factors from the karst ground collapse geospatial database and generate a sample set using a geographic detector. The evaluation model construction module is used to construct a karst ground collapse susceptibility evaluation model based on the random forest algorithm using the initial sample dataset and the sample set. The prediction module is used to divide the karst land collapse prone area into several grid cells and use the karst land collapse susceptibility evaluation model to predict the collapse probability of each grid cell. The classification module is used to import the probability value of the collapse prediction into the GIS platform and assign values ​​to the grid points according to the spatial location to obtain the karst ground collapse susceptibility index layer.

Images