A Method and System for Predicting the Spatial Scale of Earthquake Landslides Based on Graph Deep Learning

Abstract

This invention presents a method and system for predicting the spatial scale of earthquake landslides based on graph deep learning. The method divides the study area into slope units using a digital elevation model (DEM) to obtain geographic elements related to landslide hazards. It integrates the data of these geographic elements based on the slope units and constructs subgraph data using spatial connections between slope units within the grid. All subgraph data constitute a graph-structured dataset. In the subgraph, a slope unit is considered a node v, and any two nodes are considered to have a connection. The connection between nodes is represented by an edge matrix E, where all elements are 1. The landslide area within a slope unit where no landslide has occurred is set to 0. Using the graph-structured dataset, a graph Transformer model is employed to evaluate the planar area of ​​earthquake landslides within the region. This provides a practical framework and technical approach for predicting the spatial distribution of landslide hazard, particularly the scale of earthquake landslides, under extreme events.

Description

Technical Field

[0001] This invention relates to the field of geological disaster assessment and prediction technology, specifically to a method for predicting the spatial scale of earthquake landslides based on graph deep learning. Background Technology

[0002] Landslides are a common geological hazard, especially in mountainous and hilly areas with complex terrain. Extreme weather events (such as heavy rain and earthquakes) can easily trigger landslides, causing serious loss of life and property. Therefore, accurately predicting the scope and spatial scale of earthquake-induced landslides is of great significance for disaster prevention and mitigation.

[0003] The spatial scale of landslides is a crucial indicator of landslide hazard severity. However, assessing the spatial scale of all earthquake-induced landslides within a region is challenging, especially for cluster landslides triggered by extreme events. Conversely, creating a regional inventory of polygonal landslides is much easier, allowing landslide hazard assessments based on shape indices (e.g., width-to-length ratio) or other morphological measurements (e.g., area). Traditional landslide prediction models primarily model the spatial or temporal frequency of landslides using geological and meteorological data; however, these methods fail to account for the spatial scale of landslides. Furthermore, landslide sizes within a region typically exhibit a significant long-tail distribution, meaning the vast majority of slopes are either landslide-free or small, while the number of large-scale landslides is relatively small. Traditional methods exclude slopes with zero landslide area during modeling, as this could bias predictions towards smaller values ​​and underestimate large-scale landslides. However, this approach ignores the information provided by a large number of slopes with zero landslide area, failing to consider the information from all slopes globally. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to propose a graph deep learning-based method for predicting the spatial scale of earthquake landslides, providing a practical framework and technical approach for the spatial distribution of landslide hazard within a region under extreme events, particularly for predicting the scale of earthquake landslides.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention provides a method for predicting the spatial scale of earthquake landslides based on graph deep learning, the prediction method comprising the following steps:

[0007] S1. Obtain the digital elevation model of the study area and divide the study area using slope units;

[0008] S2. Obtain various geographic elements related to landslide disasters and use different integration methods to integrate each geographic element into each slope unit;

[0009] S3. Extract ramp cells by sequentially sliding a fixed-size grid. Each grid extracts multiple ramp cells. All ramp cells extracted from the same grid constitute a ramp cell subset. Each ramp cell subset is a subgraph data. In the subgraph, ramp cells are considered as nodes v. Any two nodes v... i and v j All nodes are considered to have connections, and the connections between them are represented by the edge matrix E∈R. m×m The subgraph represents the number of nodes in the subgraph data, and the elements of the edge matrix E are all 1. The characteristics of the nodes in the subgraph data are the geographic features integrated into the corresponding slope unit using different integration methods. The label data of the subgraph data nodes is the size of the planar area of ​​the landslide within the corresponding slope unit. For slope units where no landslides occur, the landslide planar area is set to 0. All subgraph data constitute a graph structure dataset of landslide influencing factors. One subgraph corresponds to one edge matrix, and there is no connection between adjacent subgraphs.

[0010] S4. Using a graph-structured dataset, employ a graph Transformer model to evaluate the planar area of ​​earthquake-induced landslides within the region.

[0011] Furthermore, the slope unit is obtained by running the r.slopunit code in the Grass GIS environment.

[0012] Preferably, the digital elevation model is an SRTM model, ASTER model, ALOS model, or Copernicus model, etc.

[0013] Furthermore, the geographic elements include: the area of ​​the slope unit, slope gradient, slope aspect, profile curvature, planar curvature, topographic location index, surface cover, stratigraphic lithology, road density, seismic intensity, and peak ground acceleration.

[0014] The integrated method includes: the mean, variance, mode, and maximum value of geographical elements within the slope unit;

[0015] For continuous variables, including slope gradient, slope aspect, profile curvature, plane curvature, and topographic location index, the mean and variance values ​​within the slope unit are used as feature integrations into the slope unit.

[0016] For categorical variables, including surface cover and stratigraphic lithology, the mode of the number of subcategories of elements within a slope unit is used as the feature to be integrated into the slope unit.

[0017] For road density, seismic intensity, and peak ground acceleration, the maximum value of the feature within the slope element is used as the feature to be integrated into the slope element.

[0018] Furthermore, the specific process of step 3 is as follows:

[0019] Step S31: Initialize the grid size, and slide the grid sequentially starting from the upper left corner of the area under study, with a sliding step size of 1 grid unit;

[0020] Step S32: Calculate the area of ​​the intersection between each slope cell and the grid of step l, and use the area ratio formula to evaluate whether each slope cell can be extracted by the grid. Use the area ratio R to determine the subset of slope cells extracted by the grid of step l. The area ratio formula is:

[0021]

[0022] Among them, u i ∩K l S(u) represents the region where the i-th slope element intersects spatially with the l-th step mesh. i ∩K l ) represents the area of ​​the region. This represents the area of ​​the i-th ramp unit;

[0023] When R≥0.5, it indicates that the slope element is extracted by the l-th step mesh; otherwise, it will not be extracted by the l-th step mesh. All slope elements extracted by the l-th step mesh constitute a subset of slope elements.

[0024] Step S33: After the mesh has traversed all slope elements in the study area, L slope element subsets are obtained and processed into L subgraph data, where L is the number of slides. The slope element is considered as node v in the subgraph data. Any two nodes v in the subgraph... i and v j Both are considered as two nodes with a connection relationship.

[0025] Furthermore, the graph Transformer model includes a node embedding layer, a position embedding module, a multi-head attention module, and a subsequent feedforward neural network.

[0026] Preferably, the graph structure dataset sequentially passes through a node embedding layer, a location embedding module, a multi-head attention module, and a subsequent feedforward neural network to output the predicted planar area of ​​the landslide.

[0027] Furthermore, the location embedding module uses the Weisfeiler-Lehman (WL) method to assign an absolute location node code to each node in the graph structure, for a subgraph node set V. i node v j Its node encoding is represented as WL(v j )∈N, the embedded node vector is represented as

[0028]

[0029] in, Represents node v j The position within the entire graph structure, based on the node's encoding WL(v) j Use the sin() and cos() functions to update the position of nodes in the subgraph within the entire graph structure; s is the total number of nodes in the graph structure dataset, d h This represents the dimension of the node features after the position is updated.

[0030] Furthermore, the graph Transformer model uses the squared error correlation region SERA function as the loss function of the graph Transformer model, and defines a correlation function. The correlation function is used to map both the true and predicted label values ​​to [0,1]. The range of the mapped true label values ​​is statistically analyzed to determine the endpoint values ​​n1 and n2, which are between 0 and 1. The area of ​​the plane is used as the x-axis, and the squared error between the mapped true and predicted values ​​is used as the y-axis. Multiple discontinuities t are set between n1 and n2, with an interval of 0.01-0.05 between adjacent t values.

[0031]

[0032] Among them, D t Indicates y i The set of values ​​greater than the discontinuity point t; SER t The squared error at the discontinuity point t. and y i Let i be the predicted value and the true value after mapping.

[0033] Furthermore, the correlation function The optional transformation type functions include: i) linear, ii) exponential, iii) logarithmic, iv) square root, and at least one of various roots; said various roots include at least one of v) square root, vi) cubic root, vii) quintic root, viiii) octet root, ix) decimal root, x) twentieth root.

[0034] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the steps of the method.

[0035] Thirdly, the present invention provides an earthquake landslide spatial scale prediction system based on graph deep learning, the system including a data processing module and a landslide area prediction module;

[0036] The data processing module divides the study area into slope units based on the digital elevation model, obtains geographic elements related to landslide disasters, integrates the data of each geographic element based on the slope units, and constructs sub-map data by utilizing the spatial connection between slope units within the grid. All sub-map data constitute a graph structure dataset.

[0037] The landslide area prediction module takes a subgraph as input and the planar area of ​​the landslide within the slope unit as output. It uses a graph structure dataset to train a graph Transformer model for predicting the area of ​​earthquake-induced landslides.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] 1. This invention uses slope elements as the basic computational unit and incorporates all slope elements into the model, meaning that slope elements with a landslide area of ​​0 are not excluded. Furthermore, the sliding mesh method is used to extract slope elements from the study area as samples of graph-structured data. This type of data structure can establish spatial dependencies within slope elements, providing richer spatial connectivity information compared to traditional structured data.

[0040] 2. This invention employs a graph Transformer model to process graph data with spatial connectivity, enabling the modeling of landslide area. By utilizing a location embedding module and a multi-head attention mechanism, it effectively captures the global spatial correlation within the landslide occurrence area, overcoming the deficiency of traditional landslide disaster spatial modeling in failing to consider the interdependencies between slopes, thereby improving the accuracy of earthquake landslide area prediction.

[0041] 3. The SERA function is used as the loss function for model training. By using the correlation function to map the label values ​​to the range of 0-1, the error of the overall distribution of the considered samples can be minimized, while taking into account the distribution characteristics of both the population and the tails. Therefore, the model architecture of this invention can estimate the planar area of ​​a landslide in the region of interest without selecting a specific statistical distribution, and can still explain the problem of sample imbalance. Attached Figure Description

[0042] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0043] Figure 1 A flowchart illustrating a graph-based deep learning-based method for predicting the spatial scale of earthquake landslides, as provided in an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of a method for extracting slope elements of a study area using a sliding fixed-size grid, as provided in an embodiment of the present invention.

[0045] In the figure, the white grid represents a fixed-size grid, with the initial grid size set to 5*5km. Multiple slope units can be extracted within each grid.

[0046] Figure 3 This is a schematic diagram of the extracted subgraph structure provided in an embodiment of the present invention, wherein the dots represent nodes, the nodes represent slope cells extracted by the mesh, the color of the nodes is mapped to the area of ​​landslides in the slope cells, and there are edges connecting any nodes.

[0047] Figure 4 This is an earthquake landslide area prediction map provided in an embodiment of the present invention. The darker the color, the larger the predicted landslide area within the slope unit. White indicates that no landslide has occurred within the slope unit, i.e., the landslide area is 0. Detailed Implementation

[0048] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0049] like Figure 1 The flowchart shown is for a graph-based deep learning-based method for predicting the spatial scale of earthquake landslides. This prediction method includes:

[0050] Step 1: Obtain the digital elevation model of the study area and divide the study area using slope units;

[0051] Step 2: Obtain geographical elements such as geological topography and hydrological environment related to landslide disasters, and use different integration methods to integrate each geographical element into the slope unit;

[0052] Step 3: Extract ramp cells by sequentially sliding a fixed-size grid. Each grid extracts multiple ramp cells. All ramp cells extracted from the same grid constitute a ramp cell subset. Each ramp cell subset is a subgraph data. In the subgraph, ramp cells are considered as nodes v. Any two nodes v... i and v j All nodes are considered to have connections, and the connections between them are represented by the edge matrix E∈R. m×m The subgraph represents the number of nodes in the subgraph data, and all elements of the edge matrix E are 1. The nodes in the subgraph data are geographic features integrated into the corresponding slope unit using different integration methods. The label data of the subgraph data nodes is the size of the planar area of ​​the landslide within the corresponding slope unit; for slope units (nodes) where no landslides occur, the planar area of ​​the landslide is set to 0. All subgraph data constitute a graph-structured dataset of landslide influencing factors. Figure 1 A matrix of edges, with no connection between adjacent subgraphs;

[0053] Step 4: Based on the graph structure dataset, use a graph Transformer model to evaluate the planar area of ​​earthquake-induced landslides within the region. The trained graph Transformer model is then used to evaluate the nonlinear mapping relationship between landslide influencing factors and the planar area of ​​the landslides, thereby predicting the spatial distribution of landslide area. The input to the graph Transformer model is a subgraph (containing the structure and corresponding features of nodes), and the output is the predicted planar area of ​​landslides within each node.

[0054] Specifically, firstly, a digital elevation model (DEM) of the study area of ​​interest can be obtained, and based on this model, the study area is divided into multiple slope units using a slope unit partitioning method. Subsequently, geographical elements related to landslide hazards, such as geological topography, hydrological environment, and land cover, can be acquired, and different integration methods are used to integrate these geographical elements into the slope units. For the extracted slope units, different sets of slope units (i.e., slope unit subsets) are created using a grid of fixed size, and these sets of slope units can then be constructed into a graph-structured dataset. Finally, the graph-structured dataset is input into a graph Transformer model for training and testing the model's performance in predicting landslide planar area within the region.

[0055] In this embodiment, optionally, in step 1, the digital elevation model is an SRTM model, ASTER model, ALOS model, Copernicus model, etc., and the spatial resolution can be selected as 12×12, 30×30, or 90×90 meters.

[0056] Optionally, in step 1, the slope unit is calculated by running the r.slopunit code in the Grass GIS environment.

[0057] In this embodiment, optionally, in step 2, the geographical elements related to landslide disasters, such as geological topography and hydrological environment, include: the area of ​​the slope unit, slope gradient, slope aspect, profile curvature, planar curvature, topographic location index, surface cover, stratigraphic lithology, road density, seismic intensity, and peak ground acceleration. Wherein, geological topography includes: the area of ​​the slope unit, slope gradient, and slope aspect; hydrological environment includes: profile curvature, planar curvature, and topographic location index.

[0058] Optionally, in step 2, the different integration methods include: the mean, variance, mode, maximum, etc. of geographical features within the slope unit.

[0059] Specifically, for continuous variable elements, such as slope gradient, slope aspect, profile curvature, plan curvature, and topographic location index, the mean and variance values ​​within the slope unit are used as feature integrations into the slope unit.

[0060] For categorical variables, such as surface cover and stratigraphic lithology, the mode of the number of subcategories of elements within a slope unit is used as the feature to be integrated into the slope unit.

[0061] For road density, seismic intensity, and peak ground acceleration features, the maximum value of the features within the slope element is used as the feature integration into the slope element.

[0062] The specific process of step 3 is as follows:

[0063] Step S31: Initialize the grid size to 5-10km. In this embodiment, it is set to 5×5km. Starting from the upper left corner of the area under study, slide the grid from left to right and from top to bottom; or slide the grid from top to bottom and from left to right. The sliding step size is 1 grid unit. Define the size of the grid, i.e., the length and width, as 5×5km.

[0064] Step S32: Calculate the area of ​​the intersection between each slope cell and the grid of step l, and use the area ratio formula to evaluate whether each slope cell can be extracted by the grid. Use the area ratio R to determine the subset of slope cells extracted by the grid of step l. The area ratio formula is:

[0065]

[0066] Among them, u i ∩K l S(u) represents the region where the i-th slope element intersects spatially with the l-th step mesh. i ∩K l ) represents the area of ​​the region. R represents the area of ​​the i-th ramp cell. When R ≥ 0.5, it indicates that the ramp cell was extracted by the l-th step grid; otherwise, it will not be extracted by the l-th step grid. All ramp cells extracted by the l-th step grid constitute a subset of ramp cells.

[0067] Step S33: After the mesh has traversed all slope elements in the study area, L slope element subsets are obtained and processed into L subgraph data, where L is the number of slides. A slope element is considered as a node v in the subgraph data. Any two nodes (slope elements) in the subgraph are considered as... i and v j Both are considered as two nodes (slope elements) with a connection relationship.

[0068] Specifically, the planar area of ​​a landslide within a slope unit can be calculated in a GIS platform by measuring the area of ​​a polygon.

[0069] In this embodiment, optionally, in step 4, the graph Transformer model includes a node embedding layer, a position embedding module, a multi-head attention module, and a subsequent feedforward neural network.

[0070] Specifically, the graph structure dataset is sequentially processed through a node embedding layer, a position embedding module, a multi-head attention module, and a subsequent feedforward neural network.

[0071] Specifically, the location embedding module uses the Weisfeiler-Lehman (WL) method to assign an absolute location node code to each node in the graph structure. Specifically, for a subgraph node set (V... i ) node v j Its node encoding is represented as WL(v j )∈N, therefore the embedded node vector is represented as

[0072]

[0073] in, Represents node v j The position within the entire graph structure, based on the node's encoding WL(v) j The sin() and cos() functions are used to update the position of nodes in the subgraph within the overall graph structure. The node features change accordingly with each position update; s is the total number of nodes in the graph structure dataset, and d... h This represents the dimension of the node features after the position is updated.

[0074] Specifically, the calculation method for attention weights in the multi-head attention module is as follows:

[0075]

[0076] Where Q, K, and V are the query matrix, key matrix, and value matrix, respectively, and d k It is the dimension of the key vector.

[0077] In this embodiment, optionally, in step 4, the graph Transformer model uses the Squared Error Related Region (SERA) function as the loss function. The purpose is to eliminate the significant long-tail distribution characteristic of extreme landslide area labels, where there are many small landslides and very few large landslides; that is, the vast majority of slope units are displayed as no landslides (landslide area is 0), and the proportion of extremely large landslides is very small, thereby avoiding modeling bias.

[0078] Define a correlation function The correlation function is used to map both the true and predicted label values ​​to [0,1]. The range of the mapped true label values ​​is statistically analyzed to determine the endpoint values ​​n1 and n2, which are between 0 and 1. The area of ​​the plane is used as the x-axis, and the squared error between the mapped true and predicted values ​​is used as the y-axis. Multiple discontinuities t are set between n1 and n2, with an interval of 0.01-0.05 between adjacent t values.

[0079] The SERA function is calculated as follows:

[0080]

[0081] Among them, D t Indicates y i The set of values ​​greater than the discontinuity point t; SER t The squared error at the discontinuity point t. and y i Let i be the predicted value and the true value after mapping.

[0082] Specifically, the SERA function, the correlation function The optional transformation function includes: i) linear, ii) exponential, iii) logarithmic, iv) square root, and various roots, including at least one of v) square root, vi) cubic root, vii) quintic root, viiii) octet root, ix) decimal root, x) decimal root, etc. A suitable correlation function should be selected based on the characteristics of the landslide; for example, for landslides with long-tailed distributions, quintic, octet, decimal, and logarithmic transformations are preferred.

[0083] In this embodiment, optionally, different correlation functions can be used during the training of the graph Transformer model. The trained graph Transformer model is validated. The specific training process for the graph Transformer model includes:

[0084] Step S41: Construct a training set and a validation set using the graph structure data;

[0085] Step S42: Standardize the training set and validation set using the standardization method respectively;

[0086] Step S43: Based on different correlation functions Different SERA loss functions are obtained to train the graph Transformer model. The training set is then input into the graph Transformer model, and the model is trained using error backpropagation and the SERA function as the loss function.

[0087] Step S44: Use the root mean square (MSE) function to evaluate the performance of each graph Transformer model on the validation set, and take the graph Transformer model with the smallest MSE value and the corresponding SERA loss function in the validation set as the optimal model for the plane area of ​​the earthquake landslide.

[0088] Step S45: Based on the performance of subsequent modeling results (such as model prediction error or prediction accuracy, the smaller the better, adjust the grid size as needed), adjust the initial grid size to (5+0.5c)×(5+0.5c)km, where c is the number of subsequent grid size adjustments, and repeat the above process to obtain a graph Transformer model that meets the error accuracy requirements, which can be used to predict the spatial distribution of landslide area.

[0089] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0090] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A method for predicting the spatial scale of earthquake landslides based on graph deep learning, characterized in that, The prediction method includes the following steps: S1. Obtain the digital elevation model of the study area and divide the study area using slope units; S2. Obtain various geographic elements related to landslide disasters and use different integration methods to integrate each geographic element into each slope unit; S3. Extract ramp cells by sequentially sliding a fixed-size grid. Each grid extracts multiple ramp cells. All ramp cells extracted from the same grid constitute a ramp cell subset. Each ramp cell subset is a subgraph data. In the subgraph, ramp cells are considered as nodes v. Any two nodes v... i and v j All nodes are considered to have connections, and the connections between them are represented by the edge matrix E∈R. m×m The subgraph represents the number of nodes in the subgraph data, and the elements of the edge matrix E are all 1. The characteristics of the nodes in the subgraph data are the geographic features integrated into the corresponding slope unit using different integration methods. The label data of the subgraph data nodes is the size of the planar area of ​​the landslide within the corresponding slope unit. For slope units where no landslides occur, the landslide planar area is set to 0. All subgraph data constitute a graph structure dataset of landslide influencing factors. One subgraph corresponds to one edge matrix, and there is no connection between adjacent subgraphs. All slope units are included in the model consideration, that is, slope units with a landslide area of ​​0 are not excluded. The specific process of step 3 is as follows: Step S31: Initialize the grid size, and slide the grid sequentially starting from the upper left corner of the area under study, with a sliding step size of 1 grid unit; Step S32: Calculate the area of ​​the intersection between each slope cell and the grid of step l, and use the area ratio formula to evaluate whether each slope cell can be extracted by the grid. Use the area ratio R to determine the subset of slope cells extracted by the grid of step l. The area ratio formula is: Among them, u i ∩K l S(u) represents the region where the i-th slope element intersects spatially with the l-th step mesh. i ∩K l ) represents the area of ​​the region. This represents the area of ​​the i-th ramp unit; When R≥0.5, it indicates that the slope element is extracted by the l-th step mesh; otherwise, it will not be extracted by the l-th step mesh. All slope elements extracted by the l-th step mesh constitute a subset of slope elements. Step S33: After the mesh has traversed all slope elements in the study area, L slope element subsets are obtained and processed into L subgraph data, where L is the number of slides. The slope element is considered as node v in the subgraph data. Any two nodes v in the subgraph... i and v j Both are considered as two nodes with a connection relationship; S4. Using a graph-structured dataset, employ a graph Transformer model to evaluate the planar area of ​​earthquake-induced landslides within the region. The graph Transformer model uses the squared error correlation region SERA function as the loss function, and defines a correlation function. The correlation function is used to map both the true and predicted label values ​​to [0,1]. The range of the mapped true label values ​​is statistically analyzed to determine the endpoint values ​​n1 and n2, which are between 0 and 1. The area of ​​the plane is used as the x-axis, and the squared error between the mapped true and predicted values ​​is used as the y-axis. Multiple discontinuities t are set between n1 and n2, with an interval of 0.01-0.05 between adjacent t values. Among them, D t Indicates y i The set of values ​​greater than the discontinuity point t; SER t The squared error at the discontinuity point t. and y i Let i be the predicted value and the true value after mapping.

2. The method for predicting the spatial scale of earthquake landslides according to claim 1, characterized in that, The slope unit is obtained by running the r.slopunit code in the Grass GIS environment; the digital elevation model is an SRTM model, ASTER model, ALOS model, or Copernicus model.

3. The method for predicting the spatial scale of earthquake landslides according to claim 1, characterized in that, The geographic elements include: area of ​​the slope unit, slope gradient, slope aspect, profile curvature, planar curvature, topographic location index, surface cover, stratigraphic lithology, road density, seismic intensity, and peak ground acceleration. The integrated method includes: the mean, variance, mode, and maximum value of geographical elements within the slope unit; For continuous variables, including slope gradient, slope aspect, profile curvature, plane curvature, and topographic location index, the mean and variance values ​​within the slope unit are used as feature integrations into the slope unit. For categorical variables, including surface cover and stratigraphic lithology, the mode of the number of subcategories of elements within a slope unit is used as the feature to be integrated into the slope unit. For road density, seismic intensity, and peak ground acceleration, the maximum value of the feature within the slope element is used as the feature to be integrated into the slope element.

4. The method for predicting the spatial scale of earthquake landslides according to claim 1, characterized in that, The graph Transformer model includes a node embedding layer, a location embedding module, a multi-head attention module, and a subsequent feedforward neural network; the graph structure dataset is passed sequentially through the node embedding layer, the location embedding module, the multi-head attention module, and the subsequent feedforward neural network to output the predicted planar area of ​​the landslide.

5. The method for predicting the spatial scale of earthquake landslides according to claim 4, characterized in that, The location embedding module uses the Weisfeiler-Lehman (WL) method to assign an absolute location node code to each node in the graph structure, for a subgraph node set V. i node v j Its node encoding is represented as WL(v j )∈N, the embedded node vector is represented as in, Represents node v j The position within the entire graph structure, based on the node's encoding WL(v) j Use the sin() and cos() functions to update the position of nodes in the subgraph within the entire graph structure; s is the total number of nodes in the graph structure dataset, d h This represents the dimension of the node features after the position is updated.

6. The method for predicting the spatial scale of earthquake landslides according to claim 1, characterized in that, The correlation function The optional transformation type functions include: i) linear, ii) exponential, iii) logarithmic, iv) square root, and at least one of various roots; said various roots include at least one of v) square root, vi) cubic root, vii) quintic root, viiii) octet root, ix) decimal root, x) twentieth root.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program can perform the steps of the method described in any one of claims 1-6.

8. A system for predicting the spatial scale of earthquake landslides based on graph deep learning, characterized in that, The system employs the steps of any one of the methods described in claims 1-6, including a data processing module and a landslide area prediction module; The data processing module divides the slope units according to the digital elevation model of the study area, obtains the geographic elements related to landslide disasters, integrates the data of each geographic element based on the slope units, and constructs sub-map data by utilizing the spatial connection between the slope units in the grid. All sub-map data constitute a graph structure dataset. The landslide area prediction module takes a subgraph as input and the planar area of ​​the landslide within the slope unit as output. It uses a graph structure dataset to train a graph Transformer model for predicting the area of ​​earthquake-induced landslides.

Images