A landslide susceptibility intelligent mapping method based on dynamic optimization

Abstract

The application discloses a landslide-prone-intelligence-mapping method based on dynamic optimization, continuously improves the precision of demand data in an iterative and dynamic optimization mode through continuous monitoring of a target region; in addition, the landslide-prone-intelligence-mapping method has less demand for landslide catalog data, and verifies the generated landslide-prone-intelligence-mapping diagram, thereby improving the precision and timeliness of the landslide-prone-intelligence-mapping diagram.

Description

Technical Field

[0001] This invention relates to the technical field of geological monitoring, and more specifically, to an intelligent mapping method for landslide susceptibility based on dynamic optimization. Background Technology

[0002] Landslide susceptibility mapping is a fundamental task in geological hazard risk assessment and disaster prevention and mitigation planning. Its core objective is to predict the likelihood of landslides occurring at different locations within the study area by analyzing the spatial correlation between historical landslide points and environmental factors (such as topography, geology, hydrology, vegetation, etc.) and to generate graded maps.

[0003] Currently, the main method for mapping landslide susceptibility is through statistical models. However, a large amount of landslide data is required when constructing statistical models. However, when the landslide sample size is small, especially in newly developed areas and sparsely populated mountainous areas, the landslide sample size is extremely small, which interferes with the accuracy of the statistical model. Summary of the Invention

[0004] To address at least one of the aforementioned technical problems, the present invention aims to provide a dynamic optimization-based intelligent mapping method for landslide susceptibility, which can improve the accuracy and timeliness of landslide susceptibility classification maps.

[0005] This invention provides a dynamically optimized intelligent mapping method for landslide susceptibility, comprising: Acquire landslide logging data and disaster-prone factor data for the target area; Using landslide points in the landslide logging data of the target area as positive samples, non-landslide points are generated as negative samples based on a preset integrated acquisition method, and a balanced training sample set is constructed. The balanced training sample set and disaster-prone factor data are input into the classifier set to generate the initial landslide susceptibility probability value of the target area; The initial landslide susceptibility probability value of the marked area is divided into intervals to obtain multiple susceptibility level interval combinations; Based on the relative distribution density of landslide points, a classification accuracy evaluation index is constructed, and the optimal interval combination is determined based on the classification accuracy evaluation index. A landslide susceptibility classification map is generated based on the optimal interval combination.

[0006] In this scheme, the step of generating a landslide susceptibility classification map specifically involves: mapping each grid cell in the target area to the susceptibility level corresponding to the optimal interval combination according to its susceptibility probability value, and rendering different levels with different colors or shadow patterns to output a landslide susceptibility classification map in grid format.

[0007] In this scheme, the step of inputting the balanced training sample set and disaster-prone factor data into the classifier set to generate the initial landslide susceptibility probability value of the target area specifically includes: The sample and disaster factor data in the balanced training sample set are divided according to a preset grid to obtain the sample and disaster factor data in grid cell s. Based on the samples and disaster risk factor data within grid cell s, determine the weight of the sample at the current time node corresponding to grid cell s. and the eigenvectors of disaster-causing factors within grid cell s ; Based on the weight of the sample at grid cell s at the current time node and the feature vector of the disaster-causing factor within grid cell s. Determine the predicted landslide s-value of the grid cell s output by the classifier. Its formula is ; Traverse all grid cells to obtain the predicted landslide susceptibility values ​​for all grid cells, which together form the initial landslide susceptibility probability values ​​for the target area.

[0008] In this scheme, the step of obtaining the weight of the sample at the current time node at grid cell s specifically includes: Based on a preset sliding time window, the samples in the balanced training sample set are updated to determine the balanced training sample set at the current time node. Set the weight of the sample at grid cell s at the first time node to... Its formula is N is the number of positive samples; If the current time point is greater than or equal to 2, then set the corresponding weight to 2. Its formula is , , This indicates the weight at the previous time point. Indicates the time decay coefficient. This represents the magnitude of change in the disaster-causing factor value from the previous time node to the current time node within grid cell s.

[0009] This plan also includes: The pregnancy factor is decomposed into a static factor and a dynamic factor; the static factor remains unchanged at different time points; the dynamic factor takes different values ​​at different time points. Let the value of the dynamic factor at time node t be set as The formula is: ;in These represent the original dynamic factor values ​​at time nodes t-1, t, and t+1, respectively.

[0010] In this solution, after obtaining the landslide logging data for the target area, the solution further includes: Set the landslide points in the landslide logging data of the target area as actual landslide points, and determine the total number of corresponding actual landslide points; If the total number of actual landslide points is less than the preset landslide point threshold, the corresponding target area is determined to be a sparse data area. Based on sparse data regions, a virtual region is constructed with actual landslide points as reference points and a preset buffer distance as the radius. Virtual landslide points are randomly generated within the virtual region, and the ratio of the number of virtual landslide points to the number of actual landslide points is less than or equal to a preset ratio threshold.

[0011] This plan also includes: Based on the fact that there are virtual landslide points in the landslide logging data of the target area, the target area is divided into multiple grid units, and the number of virtual landslide points in each grid unit is determined. The generation density of virtual landslide points in the corresponding grid cell is determined based on the number of virtual landslide points in the grid cell. The generation density of virtual landslide points in the grid cell is normalized to obtain the rejection weight for negative sample selection. Cluster analysis is performed on positive samples to generate positive sample sub-clusters; Prioritize subclusters in grid cells with smaller exclusion weights as the benchmark, and randomly generate negative samples outside a preset spatial distance until the number of negative samples equals the number of positive samples at the landslide point.

[0012] In this scheme, after generating the landslide susceptibility classification map, the following steps are also included: Each time, an actual landslide point is removed, and the remaining N-1 actual landslide points plus the virtual landslide points are used to reconstruct the balanced training sample set, and the susceptibility level of the grid cell where the removed actual landslide point is located is calculated. After traversing all actual landslide points, we obtain the set of susceptibility levels in the grid cells containing all the removed actual landslide points; The susceptibility levels in the corresponding grid cells before the actual landslide points were removed were compared with the susceptibility levels in the set of susceptibility levels in the grid cells where the actual landslide points were removed, and a set of consistency ratios was obtained. If all consistency ratios in the consistency ratio set are greater than the preset ratio threshold, then the corresponding landslide susceptibility classification map meets the requirements.

[0013] In this scheme, after generating the landslide susceptibility classification map, the following steps are also included: Based on the same grid cell, calculate the difference in susceptibility level between adjacent time nodes t-1 and t; If the difference in susceptibility levels is greater than a preset threshold, the corresponding grid cell is marked as a sudden mutation zone, and a local warning message is triggered. Based on local warning information, a comprehensive analysis of all grid cells in the target area is performed to determine the overall change index of the target area. If the overall change index of the target area is greater than the preset change index threshold, an overall change warning signal will be triggered.

[0014] One or more technical solutions proposed in this application have at least the following technical effects: 1. For areas with sparse landslide data, virtual landslide points are generated with the actual landslide points as the center, which effectively solves the problem of insufficient landslide point data in areas with sparse data; 2. The consistency and stability of the generated landslide susceptibility classification map are verified by deleting one actual landslide point at a time. In addition, the susceptibility level difference of the same grid cell at adjacent time nodes is determined by a preset level difference threshold, so as to automatically identify sudden change zones and overall changes in the target area. 3. By maximizing the classification accuracy evaluation index, the combination of susceptibility level intervals is screened to make the density of landslide points in high-susceptibility areas as high as possible and the density in low-susceptibility areas as low as possible, effectively improving the accuracy of the landslide susceptibility classification map; In summary, this invention improves the consistency, stability, and accuracy of landslide susceptibility classification maps by effectively processing landslide logging data and disaster-causing factor data in the target area. Attached Figure Description

[0015] The present invention will now be described in further detail with reference to the accompanying drawings and specific implementation methods.

[0016] Figure 1 A flowchart of a landslide susceptibility intelligent mapping method based on dynamic optimization according to the present invention is shown; Figure 2 A schematic diagram of the landslide susceptibility classification map in grid format of the present invention is shown. Detailed Implementation

[0017] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0018] Figure 1 This invention illustrates an intelligent mapping method for landslide susceptibility based on dynamic optimization.

[0019] like Figure 1 As shown, this invention discloses an intelligent mapping method for landslide susceptibility based on dynamic optimization, comprising: S101, Obtain landslide logging data and disaster-causing factor data for the target area; S102, using landslide points in the landslide logging data of the target area as positive samples, generating non-landslide points as negative samples based on a preset integrated acquisition method, and constructing a balanced training sample set; S103, input the balanced training sample set and disaster-prone factor data into the classifier set to generate the initial landslide susceptibility probability value of the target area; S104, Divide the initial landslide susceptibility probability value of the marked area into intervals to obtain multiple susceptibility level interval combinations; S105, Based on the relative distribution density of landslide points, construct a classification accuracy evaluation index, and determine the optimal interval combination based on the classification accuracy evaluation index; S106, Generate a landslide susceptibility classification map based on the optimal interval combination.

[0020] According to an embodiment of the present invention, the disaster-prone factors include the topography, geology, hydrology, and human activities of the target area; by normalizing the disaster-prone factor data, the normalized values ​​of the corresponding disaster-prone factors are determined, for example, the normalized values ​​of the disaster-prone factors are set as follows: Its formula is ,in This corresponds to the original factor value of the pregnancy and disaster factor at the current time point. and These are the minimum and maximum values ​​of the factor within the target area, respectively; the relative distribution density of landslide points is set as... Its formula is ,in This represents the relative distribution density of landslide points within the i-th susceptibility level interval. This represents the number of landslide points that fall into the i-th interval. This represents the area covered by the i-th interval. This represents the total number of landslide points within the target area. Let Q be the total area of ​​the target region; let Q be the classification accuracy evaluation index, and its formula is: Where m represents the total number of susceptibility level intervals, for example, if the susceptibility level is divided into 5 intervals, then m=5; using the boundary probability value of each level interval as the optimal variable, and the classification accuracy evaluation index as the objective function, based on the adaptive boundary adjustment algorithm, the interval boundaries are iteratively updated until the objective function value Q is greater than the preset evaluation index threshold or the preset maximum number of iterations is reached, at which point the optimal classification accuracy evaluation index is output, and the corresponding interval combination is output according to the optimal classification accuracy evaluation index, which is set as the optimal interval combination; the boundary update rule in the adaptive boundary adjustment algorithm is as follows: ,in Let be the probability value of the j-th interval boundary in the d-th iteration. Step size factor, and These represent the relative distribution densities of landslide points in the left and right sections of the boundary, respectively. When the boundary moves to the right to expand the high-density zone, it moves to the left.

[0021] According to an embodiment of the present invention, the step of generating a landslide susceptibility classification map specifically involves: mapping each grid cell within the target area to the susceptibility level corresponding to the optimal interval combination according to its susceptibility probability value, rendering different levels with different colors or shadow patterns, and outputting a landslide susceptibility classification map in grid format.

[0022] It should be noted that the optimal interval boundary is obtained through dynamic optimization, for example, by dividing it into... , , , and Five intervals, corresponding to extremely low susceptibility, low susceptibility, medium susceptibility, high susceptibility, and extremely high susceptibility; it is necessary to analyze the values ​​of each grid cell s within the target area. Based on these interval boundaries, the results are categorized into integer value levels (1-5). Then, using map-making techniques, each level is assigned an intuitive color, with higher levels corresponding to darker colors. Figure 2 As shown.

[0023] According to an embodiment of the present invention, the step of inputting the balanced training sample set and disaster-prone factor data into a classifier set to generate an initial landslide susceptibility probability value for the target area specifically includes: The sample and disaster factor data in the balanced training sample set are divided according to a preset grid to obtain the sample and disaster factor data in grid cell s. Based on the samples and disaster risk factor data within grid cell s, determine the weight of the sample at the current time node corresponding to grid cell s. and the eigenvectors of disaster-causing factors within grid cell s ; Based on the weight of the sample at grid cell s at the current time node and the feature vector of the disaster-causing factor within grid cell s. Determine the predicted landslide s-value of the grid cell s output by the classifier. Its formula is ; Traverse all grid cells to obtain the predicted landslide susceptibility values ​​for all grid cells, which together form the initial landslide susceptibility probability values ​​for the target area.

[0024] It should be noted that the specific formula for calculating the predicted landslide susceptibility value is the proportion of positive sample similarity to the total similarity, which is... ,in The weighted total similarity between grid cell s and all positive samples is given by the formula: ,in ,in Let be the Euclidean distance in the feature space. The scale parameter controls the rate at which similarity decays with distance. Let k be the feature vector of the k-th sample. For tags, where Indicates a positive sample. For negative samples, the The weighted total similarity between grid cell s and all negative samples is given by the following formula: .

[0025] It should be noted that if the number of classifiers in the classifier set is greater than or equal to 2, then multiple classifiers will be used to predict the landslide s of the balanced training sample set to obtain multiple landslide s prediction values. The multiple landslide s prediction values ​​will be averaged to obtain the average value of the landslide s prediction values. The average value of the landslide s prediction values ​​will be used to replace the multiple landslide s prediction values ​​for subsequent calculations.

[0026] According to an embodiment of the present invention, the step of obtaining the weight of the sample at the current time node at grid cell s specifically includes: Based on a preset sliding time window, the samples in the balanced training sample set are updated to determine the balanced training sample set at the current time node. Set the weight of the sample at grid cell s at the first time node to... Its formula is N is the number of positive samples; If the current time point is greater than or equal to 2, then set the corresponding weight to 2. Its formula is , , This indicates the weight at the previous time point. Indicates the time decay coefficient. This represents the magnitude of change in the disaster-causing factor value from the previous time node to the current time node within grid cell s.

[0027] It should be noted that the sample weights are continuously optimized through iteration.

[0028] According to an embodiment of the present invention, it further includes: The pregnancy factor is decomposed into a static factor and a dynamic factor; the static factor remains unchanged at different time points; the dynamic factor takes different values ​​at different time points. Let the value of the dynamic factor at time node t be set as The formula is: ;in These represent the original dynamic factor values ​​at time nodes t-1, t, and t+1, respectively.

[0029] It should be noted that in areas with seasonal rainfall or engineering disturbances, disaster-prone factors (such as rainfall, groundwater level, and vegetation cover) change over time. However, the original time series may contain noise or short-term fluctuations. Direct use of these data can lead to fluctuating susceptibility probability values, which does not conform to the geological continuity of landslide occurrence. Therefore, the values ​​of disaster-prone factors at the current time point are averaged by taking values ​​from before and after the current time.

[0030] According to an embodiment of the present invention, after obtaining the landslide logging data of the target area, the method further includes: Set the landslide points in the landslide logging data of the target area as actual landslide points, and determine the total number of corresponding actual landslide points; If the total number of actual landslide points is less than the preset landslide point threshold, the corresponding target area is determined to be a sparse data area. Based on sparse data regions, a virtual region is constructed with actual landslide points as reference points and a preset buffer distance as the radius. Virtual landslide points are randomly generated within the virtual region, and the ratio of the number of virtual landslide points to the number of actual landslide points is less than or equal to a preset ratio threshold.

[0031] It should be noted that, for example, if the preset landslide point threshold is 30, then when the total number of actual landslide points is less than 30, virtual sample expansion is performed. For example, if the preset buffer distance is 200 meters, the preset quantity ratio threshold is 5, the number of actual landslide points is 10, and the number of virtual landslide points is expanded to 20, forming a total of 30 landslide points, which just meets the conditions. The disaster-causing factor value of the virtual landslide points is obtained from the inverse distance weighted value of the actual landslide points using spatial interpolation.

[0032] Furthermore, the geological environment features at randomly generated virtual landslide points are extracted, and these features are sequentially compared and analyzed with those at actual landslide points to obtain multiple first similarity values ​​for the geological environment. If any first similarity value is greater than or equal to a preset first similarity threshold, the corresponding virtual landslide point is considered correctly set. If all first similarity values ​​are less than the preset first similarity threshold, the corresponding randomly generated virtual landslide point is deleted.

[0033] Furthermore, the distance between any two virtual landslide points is greater than or equal to a preset second distance threshold, such as a preset second distance value of 200 meters.

[0034] According to an embodiment of the present invention, it further includes: Based on the fact that there are virtual landslide points in the landslide logging data of the target area, the target area is divided into multiple grid units, and the number of virtual landslide points in each grid unit is determined. The generation density of virtual landslide points in the corresponding grid cell is determined based on the number of virtual landslide points in the grid cell. The generation density of virtual landslide points in the grid cell is normalized to obtain the rejection weight for negative sample selection. Cluster analysis is performed on positive samples to generate positive sample sub-clusters; Prioritize subclusters in grid cells with smaller exclusion weights as the benchmark, and randomly generate negative samples outside a preset spatial distance until the number of negative samples equals the number of positive samples at the landslide point.

[0035] It should be noted that, for example, if the target area is divided into 10 grid units, where grid unit 1 has 5 virtual landslide points, grid unit 2 has 2 virtual landslide points, and grid unit 3 has 0 virtual landslide points, and so on, if the total number of virtual landslide points is set to 20, then the exclusion weight for selecting negative samples in grid unit 1 is 5 / 20 = 0.25.

[0036] Furthermore, the geological environmental features of the randomly generated negative samples are extracted, and these features are sequentially compared with the geological environmental features of the actual landslide points to obtain multiple second similarity values. If any second similarity value is greater than or equal to a preset second similarity threshold, the corresponding randomly generated negative sample is deleted. If all second similarity values ​​are less than the preset second similarity threshold, the corresponding randomly generated negative sample is considered normal. The preset second similarity threshold is less than a preset first similarity threshold, for example, the preset second similarity threshold is 30%; the preset first similarity threshold is 70%.

[0037] Furthermore, the distance between negative samples generated within the same grid cell is also greater than or equal to a preset second distance threshold, and as many negative samples as possible are selected within each grid cell.

[0038] According to an embodiment of the present invention, after generating the landslide susceptibility classification map, the method further includes: Each time, an actual landslide point is removed, and the remaining N-1 actual landslide points plus the virtual landslide points are used to reconstruct the balanced training sample set, and the susceptibility level of the grid cell where the removed actual landslide point is located is calculated. After traversing all actual landslide points, we obtain the set of susceptibility levels in the grid cells containing all the removed actual landslide points; The susceptibility levels in the corresponding grid cells before the actual landslide points were removed were compared with the susceptibility levels in the set of susceptibility levels in the grid cells where the actual landslide points were removed, and a set of consistency ratios was obtained. If all consistency ratios in the consistency ratio set are greater than the preset ratio threshold, then the corresponding landslide susceptibility classification map meets the requirements.

[0039] It should be noted that since the virtual landslide points in the sparse data are generated and not real landslide points, the verification is only performed on the actual landslide points. Each time, one actual landslide point is removed, and the remaining actual landslide points and virtual landslide points are used to construct a balanced training sample set. A landslide susceptibility classification map with one missing landslide point is calculated and generated, for example, the preset proportion threshold is 70%. The consistency ratio can be determined by the ratio of the corresponding two susceptibility levels.

[0040] According to an embodiment of the present invention, after generating the landslide susceptibility classification map, the method further includes: Based on the same grid cell, calculate the difference in susceptibility level between adjacent time nodes t-1 and t; If the difference in susceptibility levels is greater than a preset threshold, the corresponding grid cell is marked as a sudden mutation zone, and a local warning message is triggered. Based on local warning information, a comprehensive analysis of all grid cells in the target area is performed to determine the overall change index of the target area. If the overall change index of the target area is greater than the preset change index threshold, an overall change warning signal will be triggered.

[0041] It should be noted that the overall change index is set to Its formula is ,in Let be the area of ​​the grid cell s, where is The difference in susceptibility level between adjacent time points t-1 and t: For example, the preset level difference threshold is set to 2, and the preset change index threshold is set to 0.5.

[0042] This invention discloses an intelligent mapping method for landslide susceptibility based on dynamic optimization. By continuously monitoring the target area, the accuracy of the required data is continuously improved through iterative and dynamic optimization. In addition, this invention requires less landslide logging data and verifies the generated landslide susceptibility classification map, thereby improving the accuracy and timeliness of the landslide susceptibility classification map.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A landslide susceptibility intelligent mapping method based on dynamic optimization, characterized in that, include: Acquire landslide logging data and disaster-prone factor data for the target area; Using landslide points in the landslide logging data of the target area as positive samples, non-landslide points are generated as negative samples based on a preset integrated acquisition method, and a balanced training sample set is constructed. The balanced training sample set and disaster-prone factor data are input into the classifier set to generate the initial landslide susceptibility probability value of the target area; The initial landslide susceptibility probability value of the marked area is divided into intervals to obtain multiple susceptibility level interval combinations; Based on the relative distribution density of landslide points, a classification accuracy evaluation index is constructed, and the optimal interval combination is determined based on the classification accuracy evaluation index. A landslide susceptibility classification map is generated based on the optimal interval combination.

2. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 1, characterized in that, The steps for generating the landslide susceptibility classification map are as follows: each grid cell in the target area is mapped to the susceptibility level corresponding to the optimal interval combination according to its susceptibility probability value, and different levels are rendered with different colors or shadow patterns to output a landslide susceptibility classification map in grid format.

3. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 1, characterized in that, The step of inputting the balanced training sample set and disaster-prone factor data into the classifier set to generate the initial landslide susceptibility probability value of the target area specifically includes: The sample and disaster factor data in the balanced training sample set are divided according to a preset grid to obtain the sample and disaster factor data in grid cell s. Based on the samples and disaster risk factor data within grid cell s, determine the weight of the sample at the current time node corresponding to grid cell s. and the eigenvectors of disaster-causing factors within grid cell s ; Based on the weight of the sample at grid cell s at the current time node and the feature vector of the disaster-causing factor within grid cell s. Determine the predicted landslide s-value of the grid cell s output by the classifier. Its formula is ; Traverse all grid cells to obtain the predicted landslide susceptibility values ​​for all grid cells, which together form the initial landslide susceptibility probability values ​​for the target area.

4. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 3, characterized in that, The steps for obtaining the weight of the sample at the current time node at grid cell s specifically include: Based on a preset sliding time window, the samples in the balanced training sample set are updated to determine the balanced training sample set at the current time node. Set the weight of the sample at grid cell s at the first time node to... Its formula is N is the number of positive samples; If the current time point is greater than or equal to 2, then set the corresponding weight to 2. Its formula is , , This indicates the weight at the previous time point. Indicates the time decay coefficient. This represents the magnitude of change in the disaster-causing factor value from the previous time node to the current time node within grid cell s.

5. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 3, characterized in that, Also includes: The pregnancy factor is decomposed into a static factor and a dynamic factor; the static factor remains unchanged at different time points; the dynamic factor takes different values ​​at different time points. Let the value of the dynamic factor at time node t be set as The formula is: ;in These represent the original dynamic factor values ​​at time nodes t-1, t, and t+1, respectively.

6. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 1, characterized in that, After obtaining the landslide logging data for the target area, the process also includes: Set the landslide points in the landslide logging data of the target area as actual landslide points, and determine the total number of corresponding actual landslide points; If the total number of actual landslide points is less than the preset landslide point threshold, the corresponding target area is determined to be a sparse data area. Based on sparse data regions, a virtual region is constructed with actual landslide points as reference points and a preset buffer distance as the radius. Virtual landslide points are randomly generated within the virtual region, and the ratio of the number of virtual landslide points to the number of actual landslide points is less than or equal to a preset ratio threshold.

7. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 6, characterized in that, Also includes: Based on the fact that there are virtual landslide points in the landslide logging data of the target area, the target area is divided into multiple grid units, and the number of virtual landslide points in each grid unit is determined. The generation density of virtual landslide points in the corresponding grid cell is determined based on the number of virtual landslide points in the grid cell. The generation density of virtual landslide points in the grid cell is normalized to obtain the rejection weight for negative sample selection. Cluster analysis is performed on positive samples to generate positive sample sub-clusters; Prioritize subclusters in grid cells with smaller exclusion weights as the benchmark, and randomly generate negative samples outside a preset spatial distance until the number of negative samples equals the number of positive samples at the landslide point.

8. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 7, characterized in that, After generating the landslide susceptibility classification map, the method further includes: Each time, an actual landslide point is removed, and the remaining N-1 actual landslide points plus the virtual landslide points are used to reconstruct the balanced training sample set, and the susceptibility level of the grid cell where the removed actual landslide point is located is calculated. After traversing all actual landslide points, we obtain the set of susceptibility levels in the grid cells containing all the removed actual landslide points; The susceptibility levels in the corresponding grid cells before the actual landslide points were removed were compared with the susceptibility levels in the set of susceptibility levels in the grid cells where the actual landslide points were removed, and a set of consistency ratios was obtained. If all consistency ratios in the consistency ratio set are greater than the preset ratio threshold, then the corresponding landslide susceptibility classification map meets the requirements.

9. The intelligent mapping method for landslide susceptibility based on dynamic optimization according to claim 8, characterized in that, After generating the landslide susceptibility classification map, the method further includes: Based on the same grid cell, calculate the difference in susceptibility level between adjacent time nodes t-1 and t; If the difference in susceptibility levels is greater than a preset threshold, the corresponding grid cell is marked as a sudden mutation zone, and a local warning message is triggered. Based on local warning information, a comprehensive analysis of all grid cells in the target area is performed to determine the overall change index of the target area. If the overall change index of the target area is greater than the preset change index threshold, an overall change warning signal will be triggered.

Images