Slope stability assessment method and system of graph convolution integrated network based on attention mechanism, terminal and storage medium

Abstract

The invention belongs to the technical field of slope stability analysis, and discloses a slope stability evaluation method and system of a graph convolution integrated network based on an attention mechanism, a terminal and a storage medium, and the method comprises the steps: obtaining initial InSAR ground deformation point data and landslide point data, and carrying out the preprocessing, obtaining an InSAR ground deformation point and node feature sample data set of an adjacent area of the landslide point; constructing an initial attention mechanism-based graph convolution integration network, and training and testing the initial attention mechanism-based graph convolution integration network by using the node feature sample data set to obtain an attention mechanism-based graph convolution integration network; and slope stability evaluation is carried out by using the attention mechanism-based graph convolution integrated network. According to the method, for multi-source heterogeneous slope stability sample data, the spatial correlation between the ground deformation points of the global region is considered, the landslide hidden danger of the landslide adjacent region is considered, and high-precision slope stability evaluation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of slope stability analysis, and particularly relates to a slope stability assessment method, system, terminal, and computer-readable storage medium based on an attention mechanism-based graph convolution integration network. Background Art

[0002] Landslides caused by slope instability have a significant impact on public infrastructure and the ecological environment. Therefore, slope stability assessment is of great significance for preventing and reducing landslides.

[0003] With the development of artificial intelligence technology, various existing machine learning methods are applied to slope stability assessment. For example, a single neural network uses slope stability influencing factors as input data and slope safety factor values as output data. Although it has good performance effects, almost all of them predict the slope safety factor values of individual slope points, without considering the influence of the spatial correlation of ground deformation between all slope points in the study area on the surrounding area, resulting in ignoring the potential slope instability risks in adjacent areas, that is, not considering the influence of the spatial correlation between ground deformation points in the global area on the slope stability prediction results. Moreover, a single machine learning network is suitable for homogeneous data structures and not suitable for multi-source heterogeneous slope stability sample data.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0005] The main purpose of the present invention is to provide a slope stability assessment method, system, terminal, and computer-readable storage medium based on an attention mechanism-based graph convolution integration network, aiming to solve the problems that the existing slope stability assessment method based on a single network is not suitable for multi-source heterogeneous landslide data and does not consider the influence of the spatial correlation between ground deformation points in the global area on slope stability.

[0006] To achieve the above invention purpose, the present invention provides a slope stability assessment method based on an attention mechanism-based graph convolution integration network. The slope stability assessment method based on an attention mechanism-based graph convolution integration network includes: Obtain initial InSAR ground deformation point data and landslide point data, and preprocess the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the adjacent area of the landslide point and a node feature sample data set based on the InSAR ground deformation points; Construct an initial attention mechanism-based graph convolution integration network, and use the node feature sample data set to train and test the initial attention mechanism-based graph convolution integration network to obtain an attention mechanism-based graph convolution integration network with an error index meeting the assessment accuracy requirements; Obtain the InSAR ground deformation point data to be evaluated, and input the InSAR ground deformation point data to be evaluated into the graph convolutional integration network based on the attention mechanism to output the slope stability evaluation result corresponding to the InSAR ground deformation point data to be evaluated.

[0007] Optionally, the obtaining of the initial InSAR ground deformation point data and landslide point data, and the preprocessing of the initial InSAR ground deformation point data and the landslide point data to obtain the InSAR ground deformation points in the adjacent area of the landslide point and the node feature sample data set based on the InSAR ground deformation points specifically includes: Obtain the initial InSAR ground deformation point data, landslide point data, publicly available multi-source landslide inducing factor data, ground elevation data and slope data of the research area; According to the initial InSAR ground deformation point data, the landslide point data, the publicly available multi-source landslide inducing factor data, the ground elevation data and the slope data, through spatial correlation analysis, spatial multi-value extraction analysis, importance ranking and safety factor value calculation, obtain the associated slope stability influence factor sequence in the adjacent area of the landslide point and the InSAR ground deformation points with safety factor value attributes; Based on the associated slope stability influence factor sequence and the InSAR ground deformation points with safety factor value attributes, through spatial correlation calculation, construct an undirected network graph, and use the undirected network graph to obtain the node feature sample data set based on the InSAR ground deformation points.

[0008] Optionally, the obtaining of the associated slope stability influence factor sequence in the adjacent area of the landslide point and the InSAR ground deformation points with safety factor value attributes according to the initial InSAR ground deformation point data, the landslide point data, the publicly available multi-source landslide inducing factor data, the ground elevation data and the slope data through spatial correlation analysis, spatial multi-value extraction analysis, importance ranking and safety factor value calculation specifically includes: According to the initial InSAR ground deformation point data and the landslide point data, through spatial correlation analysis, establish the attribute association between the landslide point and the initial InSAR ground deformation point to obtain the InSAR ground deformation points associated with the landslide attribute, where the InSAR ground deformation points are the initial InSAR ground deformation points in the adjacent area of the landslide point; According to the InSAR ground deformation points and the publicly available multi-source landslide inducing factor data, through spatial multi-value extraction analysis, obtain multiple slope stability influence factors; The importance index of each of the slope stability influence factors is analyzed using the information gain algorithm, and the slope stability influence factors are sorted in descending order according to the importance index of each of the slope stability influence factors to obtain a sequence of slope stability influence factors; Based on the ground elevation data and the slope data corresponding to the InSAR ground deformation points, the safety factor value of the InSAR ground deformation points is calculated using the finite element method; Based on the InSAR ground deformation points, the sequence of associated slope stability influence factors in the adjacent area of the landslide point and the InSAR ground deformation points with the safety factor value attribute are obtained; Among them, the InSAR ground deformation point data corresponding to the InSAR ground deformation point with the sequence of associated slope stability influence factors and the safety factor value attribute includes the safety factor value, the deformation rate, the sequence of slope stability influence factors, and the safety factor value.

[0009] Optionally, based on the InSAR ground deformation points with the sequence of associated slope stability influence factors and the safety factor value attribute, an undirected network graph is constructed through spatial correlation calculation, and a node feature sample data set based on the InSAR ground deformation points is obtained using the undirected network graph. Specifically, it includes: Based on the InSAR ground deformation points with the sequence of associated slope stability influence factors and the safety factor value attribute, spatial correlation data based on the InSAR ground deformation points is obtained through spatial correlation calculation. Among them, the spatial correlation data includes spatial correlation values, and the spatial correlation values are calculated using the Gaussian similarity function: =exp ; Among them, represents the spatial correlation value between the th InSAR ground deformation point and the th InSAR ground deformation point, represents the Euclidean distance between the th InSAR ground deformation point and the th InSAR ground deformation point, represents the standard deviation of the Euclidean distance between the th InSAR ground deformation point and the th InSAR ground deformation point, and exp represents the exponential function; ​Construct an undirected network graph based on the InSAR ground deformation points according to the associated slope stability influence factor sequence and the safety factor value attribute, and the spatial correlation data based on the InSAR ground deformation points. Use the InSAR ground deformation points as the nodes of the undirected network graph, use the deformation rate of the InSAR ground deformation points and the slope stability influence factor sequence as the deformation rate and slope stability influence factor sequence of the corresponding nodes, and use the spatial correlation value between the InSAR ground deformation points as the spatial similarity value and node edge weight between the corresponding nodes to obtain the undirected network graph , where \(v\) represents the set of nodes in the undirected network graph \(g\), represents the edge dataset, and \(A\) represents the node edge weight matrix; Use the undirected network graph to obtain a node feature sample dataset, where the node feature sample dataset includes input features corresponding to nodes, and the input features include the deformation rate and the slope stability influence factor sequence: ; ; where, represents the node feature sample dataset, represents the input feature corresponding to the first node, represents the input feature corresponding to the second node, represents the th input feature corresponding to the th node, represents the th input feature corresponding to the th node, represents the total number of nodes, represents the deformation rate corresponding to the th node, represents the first slope stability influence factor corresponding to the th node, represents the second slope stability influence factor corresponding to the th node, represents the th slope stability influence factor corresponding to the

[0010] Optionally, construct an initial graph convolutional integration network based on the attention mechanism, and use the node feature sample data set to train and test the initial graph convolutional integration network based on the attention mechanism to obtain a graph convolutional integration network based on the attention mechanism whose error index meets the evaluation accuracy requirements, specifically including: Construct an initial graph convolutional integration network based on the attention mechanism, where the initial graph convolutional integration network based on the attention mechanism includes an initial graph convolutional neural network based on the attention mechanism and an initial hybrid ensemble learning network. The initial graph convolutional neural network based on the attention mechanism includes an attention mechanism layer and a graph convolutional neural network, and the initial hybrid ensemble learning network includes a sub-network model layer and a meta-machine learning classifier; Input the node feature sample data set into the initial graph convolutional neural network based on the attention mechanism, and the initial graph convolutional neural network based on the attention mechanism extracts spatial correlation features from the node feature sample data set to obtain spatial correlation features corresponding to the node feature sample data set; Input the spatial correlation features corresponding to the node feature sample data set into the initial hybrid ensemble learning network, and the initial hybrid ensemble learning network performs slope stability evaluation on the spatial correlation features to obtain a slope stability evaluation result corresponding to the node feature sample data set, where the slope stability evaluation result corresponding to the node feature sample data set is the slope stability evaluation result corresponding to the InSAR ground deformation point; Compare the slope stability evaluation result corresponding to the InSAR ground deformation point with the slope stability calculation result corresponding to the safety factor value in the InSAR ground deformation point data, and perform iterative training and testing on the initial graph convolutional integration network based on the attention mechanism according to the comparison result until the error index of the initial graph convolutional integration network based on the attention mechanism meets the evaluation accuracy requirements to obtain a graph convolutional integration network based on the attention mechanism; Among them, the error index includes mean absolute error, mean square error, and root mean square error.

[0011] Optionally, the step of inputting the node feature sample data set into the initial graph convolutional neural network based on the attention mechanism, and the initial graph convolutional neural network based on the attention mechanism extracts spatial correlation features from the node feature sample data set to obtain spatial correlation features corresponding to the node feature sample data set, specifically includes: Input the node feature sample data set into the attention mechanism layer. The activation function of the attention mechanism layer calculates the weight value of each node based on the deformation rate of each node and the slope stability influence factor sequence in the node feature sample data set, and calculates the attention matrix based on the weight value of each node and the input features: ; Wherein, represents the attention matrix, represents the activation function of the attention mechanism layer, represents the weight value of the th node, represents the th node bias; Perform normalization processing on the attention matrix to obtain the normalized attention matrix: ; Wherein, represents the component of the normalized attention matrix ', represents the component of the attention matrix ; Multiply the normalized attention matrix by the node edge weight matrix to obtain the attention feature set ; Input the attention feature set into the graph convolutional neural network. Each layer of sub-graph convolutional neural network in the graph convolutional neural network outputs spatial correlation features based on the attention feature set and the node edge weight matrix, and uses the spatial correlation features as the input data for the next layer of sub-graph convolutional neural network until the spatial correlation features output by the last layer of sub-graph convolutional neural network are obtained. Among them, the spatial correlation features output by the last layer of sub-graph convolutional neural network are the spatial correlation features corresponding to the node feature sample data set: = A+ ; ; ; ; Wherein, represents the intermediate node edge weight matrix, represents the Laplace equation matrix coefficient, represents the identity matrix of represents the new node edge weight matrix, represents The degree matrix and are respectively the spatial correlation features output by the sub-graph convolutional neural network of the -th layer and the sub-graph convolutional neural network of the -th layer in the graph convolutional neural network. and are respectively the initial activation function and the initial weight value of the graph convolutional neural network. and are respectively the activation function and the weight value of the sub-graph convolutional neural network of the -th layer in the graph convolutional neural network. and are respectively the activation function and the weight value of the sub-graph convolutional neural network of the -th layer in the graph convolutional neural network.

[0012] Optionally, when inputting the spatial correlation features corresponding to the node feature sample dataset into the initial hybrid ensemble learning network, the initial hybrid ensemble learning network performs slope stability evaluation on the spatial correlation features to obtain the slope stability evaluation result corresponding to the node feature sample dataset, specifically including: Inputting the spatial correlation features corresponding to the node feature sample dataset into the sub-network model layer, and multiple basic neural network models in the sub-network model layer perform preliminary slope stability evaluation based on the spatial correlation features and output multiple preliminary slope stability evaluation results; Inputting multiple preliminary slope stability evaluation results into the meta-machine learning classifier, and the meta-machine learning classifier performs logistic regression processing based on multiple preliminary slope stability evaluation results and outputs the safety factor value of the InSAR ground deformation point corresponding to the node feature sample dataset; When the safety factor value of the InSAR ground deformation point is greater than the preset threshold, it is determined that the slope stability evaluation result of the InSAR ground deformation point is stable; When the safety factor value of the InSAR ground deformation point is not greater than the preset threshold, it is determined that the slope stability evaluation result of the InSAR ground deformation point is unstable.

[0013] To achieve the above object of the invention, the present invention also provides a slope stability evaluation system based on an attention mechanism graph convolutional integration network, and the slope stability evaluation system based on the attention mechanism graph convolutional integration network includes: Sample dataset construction module: used to obtain initial InSAR ground deformation point data and landslide point data, and preprocess the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the adjacent area of the landslide point and a node feature sample dataset based on the InSAR ground deformation points; Network construction module: used to construct an initial graph convolutional integration network based on the attention mechanism, train and test the initial graph convolutional integration network based on the attention mechanism using the node feature sample dataset, and obtain a graph convolutional integration network based on the attention mechanism whose error index meets the evaluation accuracy requirements; Slope stability evaluation module: used to obtain InSAR ground deformation point data to be evaluated, input the InSAR ground deformation point data to be evaluated into the graph convolutional integration network based on the attention mechanism, and output the slope stability evaluation result corresponding to the InSAR ground deformation point data to be evaluated.

[0014] To achieve the above invention purpose, the present invention also provides a terminal, the terminal includes: a memory, a processor, and a slope stability evaluation program of a graph convolutional integration network based on the attention mechanism stored on the memory and operable on the processor. When the slope stability evaluation program of the graph convolutional integration network based on the attention mechanism is executed by the processor, it implements the steps of the slope stability evaluation method of the graph convolutional integration network based on the attention mechanism as described above.

[0015] To achieve the above invention purpose, the present invention also provides a computer-readable storage medium, the computer-readable storage medium stores a slope stability evaluation program of a graph convolutional integration network based on the attention mechanism. When the slope stability evaluation program of the graph convolutional integration network based on the attention mechanism is executed by a processor, it implements the steps of the slope stability evaluation method of the graph convolutional integration network based on the attention mechanism as described above.

[0016] In the present invention, initial InSAR (Interferometric Synthetic Aperture Radar) ground deformation point data and landslide point data are acquired, and the initial InSAR ground deformation point data and the landslide point data are preprocessed to obtain InSAR ground deformation points in the adjacent area of the landslide point and a node feature sample data set based on the InSAR ground deformation points; an initial graph convolutional integration network based on the attention mechanism is constructed, and the initial graph convolutional integration network based on the attention mechanism is trained and tested by using the node feature sample data set to obtain a graph convolutional integration network based on the attention mechanism whose error index meets the evaluation accuracy requirement; the to-be-evaluated InSAR ground deformation point data is acquired, and the to-be-evaluated InSAR ground deformation point data is input into the graph convolutional integration network based on the attention mechanism, and a slope stability evaluation result corresponding to the to-be-evaluated InSAR ground deformation point data is output. The present invention combines the performance of a deep learning network and an ensemble learning network, faces multi-source heterogeneous slope stability sample data, takes into account the spatial correlation between InSAR ground deformation points in the global area, and realizes high-precision slope stability evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of a preferred embodiment of the slope stability evaluation method of the graph convolutional integration network based on the attention mechanism of the present invention; Figure 2 is a structural diagram of a preferred embodiment of the slope stability evaluation method of the graph convolutional integration network based on the attention mechanism of the present invention; Figure 3 is a framework diagram of the AGCN-HEN network of the present invention; Figure 4 is a framework diagram of the AGCN network of the present invention; Figure 5 is a framework diagram of the HEN network of the present invention; Figure 6 is a structural diagram of a preferred embodiment of the slope stability evaluation system of the graph convolutional integration network based on the attention mechanism of the present invention; Figure 7 is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] In order to make the objectives, technical solutions and advantages of the present invention clearer and more definite, the present invention will be further described in detail below with reference to the accompanying drawings and by way of examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] Landslides caused by slope instability have a significant impact on public infrastructure and the ecological environment. Therefore, slope stability assessment is of great significance for landslide prevention and mitigation.

[0020] At present, many methods have been used for slope stability assessment. For example, the analytical method constructs a mathematical model of slope stability and uses relevant analytical formulas for solution, but it is not applicable to the stability assessment of complex and large-scale slopes. In addition, the finite element method and the finite difference method are commonly used numerical analysis methods. This method regards slope stability assessment as a discrete finite-dimensional mathematical calculation problem that can obtain an approximate solution and can be used for the stability assessment of complex slopes. However, this method takes a long time to calculate, requires field measurements to determine relevant parameters, and accurate boundary conditions are needed for high-precision slope stability results, which are difficult to obtain in practical applications. InSAR technology is one of the commonly used remote sensing means for monitoring surface deformation, and it assesses slope stability through the temporal variation of ground displacement. However, in mountainous jungle areas with large undulations, there are problems of geometric distortion in the InSAR images obtained by this technology, including foreshortening, layover, and shadow. Although InSAR technology can be used for large-scale slope stability assessment, accurate assessment results need to be combined with topographic and geological data, and even field investigations to comprehensively judge the slope instability state of the area.

[0021] With the development of artificial intelligence technology, various existing machine learning methods have been applied to slope stability assessment, such as single neural networks: random forest, logistic regression, support vector machine, artificial neural network, etc. These network models use slope stability influencing factors as input data and the slope safety factor value FOS (factor of safety) as output data. Although they have achieved good performance, almost all of them predict the slope safety factor value of a single slope point and do not consider the influence of the spatial correlation of ground deformation between all slope points in the study area on the surrounding area, resulting in the potential slope instability risk (i.e., landslide hazard) in the adjacent area being ignored, that is, the influence of the spatial correlation between ground deformation points in the global area on the slope stability prediction result is not considered. Moreover, a single machine learning network is suitable for homogeneous data structures and is not applicable to the analysis of multi-source heterogeneous slope stability sample data (including multi-source heterogeneous landslide data).

[0022] Since a single machine learning network is applicable to homogeneous data structures, it is necessary to integrate the performance of multiple single networks for the analysis of multi-source heterogeneous slope stability sample data. Given that the existing deep learning and ensemble learning networks have better performance in slope stability assessment than single machine learning networks, the present invention integrates the performance of deep learning networks and ensemble learning networks, faces multi-source heterogeneous slope stability sample data, and is guided by InSAR ground deformation, taking into account the spatial correlation between ground deformation points in the global region, and proposes an attention-based graph convolutional ensemble network AGCN-HEN (attention-graph convolutional network-hybrid ensemble learning network) for slope stability assessment, obtains initial InSAR ground deformation point data and landslide point data, and preprocesses the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the adjacent area of the landslide point and a node feature sample data set based on the InSAR ground deformation points; constructs an initial attention-based graph convolutional ensemble network, uses the node feature sample data set to train and test the initial attention-based graph convolutional ensemble network, and obtains an attention-based graph convolutional ensemble network with an error index meeting the assessment accuracy requirements; obtains InSAR ground deformation point data to be evaluated, inputs the InSAR ground deformation point data to be evaluated into the attention-based graph convolutional ensemble network, and outputs the slope stability assessment result corresponding to the InSAR ground deformation point data to be evaluated. The present invention integrates the performance of deep learning networks and ensemble learning networks, faces multi-source heterogeneous slope stability sample data, takes into account the spatial correlation between InSAR ground deformation points in the global region, and realizes high-precision slope stability assessment.

[0023] The following further describes the application content by describing the embodiments in conjunction with the accompanying drawings.

[0024] A preferred embodiment of the slope stability assessment method of the attention-based graph convolutional ensemble network of the present invention is as Figure 1 and Figure 2 shown, and specifically includes: S1. Obtain initial InSAR ground deformation point data and landslide point data, and preprocess the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the adjacent area of the landslide point and a node feature sample data set based on the InSAR ground deformation points.

[0025] In an implementation of this embodiment, the steps of obtaining the initial InSAR ground deformation point data and landslide point data, and preprocessing the initial InSAR ground deformation point data and the landslide point data to obtain the InSAR ground deformation points in the adjacent area of the landslide point and the node feature sample data set based on the InSAR ground deformation points specifically include: Obtain the initial InSAR ground deformation point data, landslide point data, publicly available multi-source landslide inducing factor data, ground elevation data, and slope data of the research area; According to the initial InSAR ground deformation point data, the landslide point data, the publicly available multi-source landslide inducing factor data, the ground elevation data, and the slope data, through spatial correlation analysis, spatial multi-value extraction analysis, importance ranking, and safety factor value calculation, obtain the sequence of associated slope stability influencing factors in the adjacent area of the landslide point and the InSAR ground deformation points with safety factor value attributes; Based on the InSAR ground deformation points with the sequence of associated slope stability influencing factors and safety factor value attributes, through spatial correlation calculation, construct an undirected network graph, and use the undirected network graph to obtain the node feature sample data set based on the InSAR ground deformation points.

[0026] In an implementation of this embodiment, the steps of obtaining the sequence of associated slope stability influencing factors in the adjacent area of the landslide point and the InSAR ground deformation points with safety factor value attributes according to the initial InSAR ground deformation point data, the landslide point data, the publicly available multi-source landslide inducing factor data, the ground elevation data, and the slope data, through spatial correlation analysis, spatial multi-value extraction analysis, importance ranking, and safety factor value calculation specifically include: According to the initial InSAR ground deformation point data and the landslide point data, through spatial correlation analysis, establish the attribute association between the landslide point and the initial InSAR ground deformation point, and obtain the InSAR ground deformation points associated with the landslide attribute, where the InSAR ground deformation points are the initial InSAR ground deformation points in the adjacent area of the landslide point; According to the InSAR ground deformation points and the publicly available multi-source landslide inducing factor data, through spatial multi-value extraction analysis, obtain multiple slope stability influencing factors; Adopt the information gain algorithm to analyze the importance index of each slope stability influencing factor, and arrange each slope stability influencing factor in descending order according to the importance index of each slope stability influencing factor to obtain the sequence of slope stability influencing factors; According to the ground elevation data and the slope data corresponding to the InSAR ground deformation points, use the finite element method to calculate the safety factor value of the InSAR ground deformation points. Based on the InSAR ground deformation points, the sequence of slope stability influence factors, and the safety factor values, obtain the InSAR ground deformation points associated with the sequence of slope stability influence factors and the safety factor value attributes in the adjacent area of the landslide point; Among them, the InSAR ground deformation point data corresponding to the InSAR ground deformation point associated with the sequence of slope stability influence factors and the safety factor value attributes includes the safety factor value, the deformation rate, the sequence of slope stability influence factors, and the safety factor value.

[0027] In an implementation manner of this embodiment, based on the InSAR ground deformation point associated with the sequence of slope stability influence factors and the safety factor value attributes, through spatial correlation calculation, construct an undirected network graph, and use the undirected network graph to obtain a node feature sample data set based on the InSAR ground deformation point, specifically including: Based on the InSAR ground deformation point associated with the sequence of slope stability influence factors and the safety factor value attributes, through spatial correlation calculation, obtain the spatial correlation data based on the InSAR ground deformation point. Among them, the spatial correlation data includes spatial correlation values, and the spatial correlation values are calculated through a Gaussian similarity function: =exp ; Among them, represents the spatial correlation value between the th InSAR ground deformation point and the th InSAR ground deformation point, represents the Euclidean distance between the th InSAR ground deformation point and the th InSAR ground deformation point, represents the Euclidean distance between the th InSAR ground deformation point and the th InSAR ground deformation point standard deviation, and exp represents the exponential function; According to the InSAR ground deformation point associated with the sequence of slope stability influence factors and the safety factor value attributes and the spatial correlation data based on the InSAR ground deformation point, construct an undirected network graph, use the InSAR ground deformation point as the node of the undirected network graph, use the deformation rate and the sequence of slope stability influence factors of the InSAR ground deformation point as the deformation rate and the sequence of slope stability influence factors of the corresponding node, and use the spatial correlation value between the InSAR ground deformation points as the spatial similarity value and the node edge weight between the corresponding nodes to obtain the undirected network graph , where v represents in the undirected network graph g A set of nodes represents the edge data set, and A represents the node-edge weight matrix; Using the undirected network graph to obtain a node feature sample data set, where the node feature sample data set includes input features corresponding to nodes, and the input features include the deformation rate and the slope stability influence factor sequence: ; ; wherein, represents the node feature sample data set, represents the input feature corresponding to the first node, represents the input feature corresponding to the second node, represents the input feature corresponding to the th node, represents the input feature corresponding to the th node, represents the total number of nodes, represents the deformation rate corresponding to the th node, represents the first slope stability influence factor corresponding to the th node, represents the second slope stability influence factor corresponding to the th node, represents the th slope stability influence factor corresponding to the th node, and

[0028] represents the total number of slope stability influence factors in the slope stability influence factor sequence. Specifically, in order to extract the spatial correlation characteristics between ground deformation points, the present invention needs to construct a slope stability sample data set (i.e., node feature sample data set) based on InSAR points (i.e., InSAR ground deformation points), including a slope stability influence sequence factor data set (i.e., slope stability influence factor sequence) and spatial correlation data (i.e., spatial correlation data) based on InSAR ground deformation points. The specific process is as follows: Construct a dataset of sequence factors affecting slope stability based on InSAR points: First, according to the obtained landslide points, InSAR ground deformation points (original InSAR ground deformation points, i.e., initial InSAR ground deformation points), and publicly available multi-source landslide inducing factors (such as slope, slope height, and soil unit weight, etc.) data, use the spatial join method to establish the attribute association between the landslide points and InSAR ground deformation points in the study area, and obtain the InSAR ground deformation points in the adjacent area of the landslide points (i.e., InSAR ground deformation points associated with landslide attributes / InSAR ground deformation points with potential landslide hazards). Then, based on this data, use the spatial multi-value extraction analysis tool of the ArcGIS (Arc Geographic Information System) platform to obtain the InSAR ground deformation point data associated with the attributes of slope stability influencing factors (which can be understood as obtaining the slope stability influencing factor data of InSAR ground deformation points). Since the Information Gain (IG) algorithm can measure the contribution degree of variable features to the target classification result, the present invention uses the IG algorithm to analyze the importance index of each of the above slope stability influencing factors, eliminates the influencing factors with extremely small influence on slope stability, and then arranges them in descending order according to their importance index to obtain a dataset of sequence factors affecting slope stability (i.e., a sequence of slope stability influencing factors), which is used as the input data of the slope stability assessment model constructed by the present invention. And, taking the slope safety factor value FOS extracted based on the above InSAR ground deformation points as the output data of the model, the present invention uses the ground elevation and slope data of the study area, and uses the finite element method in SAGA GIS (System for Automated Geoscientific Analyse) to calculate the safety factor value FOS of the study area, where the study area with an FOS value less than or equal to 1 is classified as an unstable state, i.e., FOS <= 1, which is represented as an unstable state and encoded as 0; otherwise, it is a stable state and encoded as 1.

[0029] Constructing spatial correlation data based on InSAR ground deformation points: To consider the influence of the spatial correlation characteristics of ground deformation points on slope stability, it is necessary to establish spatial correlation data. To represent the spatial correlation characteristics of InSAR ground deformation points throughout the entire study area, an undirected network graph (GGDN, global ground deformation network) structure is constructed based on the above-obtained sequence of associated slope stability influence factors and the InSAR ground deformation points with FOS safety coefficient value attributes, which is used to describe the spatial topological structure relationship between InSAR ground deformation points. Each InSAR ground deformation point in the undirected network graph is regarded as a graph node, and the edge dataset reflects the spatial connectivity between nodes. Each value in the node-edge weight matrix represents the weight value (edge weight value) between nodes. It should be noted that the edge dataset consists of spatial similarity values between two nodes, and these similarity values represent the connection weights between the positions of ground deformation points. The greater the weight, the greater the spatial correlation it represents, that is, spatial similarity value = edge weight value = spatial correlation value.

[0030] The present invention obtains a large amount of ground deformation point data based on InSAR technology, and proposes a construction method for a slope stability sample dataset based on InSAR points, which solves the problem that the sparsity of landslide point data cannot meet the requirements of a large number of training samples required for deep learning; at the same time, aiming at the problem that not all InSAR ground deformation points will cause landslides, based on the first law of geography, a sample dataset of InSAR ground deformation points with potential landslide hazards is constructed, providing a data basis for slope stability assessment.

[0031] S2. Construct an initial graph convolutional ensemble network based on the attention mechanism, and use the node feature sample dataset to train and test the initial graph convolutional ensemble network based on the attention mechanism to obtain a graph convolutional ensemble network based on the attention mechanism with an error index meeting the evaluation accuracy requirements.

[0032] In an implementation manner of this embodiment, the constructing an initial graph convolutional ensemble network based on the attention mechanism, using the node feature sample dataset to train and test the initial graph convolutional ensemble network based on the attention mechanism to obtain a graph convolutional ensemble network based on the attention mechanism with an error index meeting the evaluation accuracy requirements specifically includes: Construct an initial graph convolutional ensemble network based on the attention mechanism, where the initial graph convolutional ensemble network based on the attention mechanism includes an initial graph convolutional neural network based on the attention mechanism and an initial hybrid ensemble learning network. The initial graph convolutional neural network based on the attention mechanism includes an attention mechanism layer and a graph convolutional neural network, and the initial hybrid ensemble learning network includes a sub-network model layer and a meta-machine learning classifier; Input the node feature sample data set into the initial graph convolutional neural network based on the attention mechanism. The initial graph convolutional neural network based on the attention mechanism extracts spatial correlation features from the node feature sample data set to obtain the spatial correlation features corresponding to the node feature sample data set; Input the spatial correlation features corresponding to the node feature sample data set into the initial hybrid ensemble learning network. The initial hybrid ensemble learning network performs slope stability evaluation on the spatial correlation features to obtain the slope stability evaluation result corresponding to the node feature sample data set. Among them, the slope stability evaluation result corresponding to the node feature sample data set is the slope stability evaluation result corresponding to the InSAR ground deformation point; Compare the slope stability evaluation result corresponding to the InSAR ground deformation point with the slope stability calculation result corresponding to the safety factor value in the InSAR ground deformation point data. Iteratively train and test the initial graph convolutional integration network based on the attention mechanism according to the comparison result until the error index of the initial graph convolutional integration network based on the attention mechanism meets the evaluation accuracy requirement to obtain the graph convolutional integration network based on the attention mechanism; Among them, the error index includes mean absolute error, mean square error, and root mean square error.

[0033] In an implementation manner of this embodiment, the inputting the node feature sample data set into the initial graph convolutional neural network based on the attention mechanism, and the initial graph convolutional neural network based on the attention mechanism extracting spatial correlation features from the node feature sample data set to obtain the spatial correlation features corresponding to the node feature sample data set specifically includes: Input the node feature sample data set into the attention mechanism layer. The activation function of the attention mechanism layer calculates the weight value of each node according to the deformation rate and slope stability influence factor sequence of each node in the node feature sample data set, and calculates the attention matrix based on the weight value and input features of each node: ; Among them, represents M an attention matrix of dimensions, represents the activation function of the attention mechanism layer, represents the weight value of the th node, represents the deviation of the th node; It should be noted that the activation function of the attention mechanism layer is ReLU (Rectified Linear Unit, linear rectifier unit); Normalize the attention matrix to obtain a normalized attention matrix: ; where represents the component of the normalized attention matrix '; represents the component of the attention matrix , indicating the degree of spatial correlation between node and node ; ' is the result of normalizing matrix , aiming to ensure that the sum of weights of all nodes is 1; Multiply the normalized attention matrix by the node-edge weight matrix to obtain an attention feature set ; Input the attention feature set into the graph convolutional neural network. Each sub-graph convolutional neural network layer in the graph convolutional neural network outputs spatial correlation features based on the attention feature set and the node-edge weight matrix, and uses the spatial correlation features as the input data for the next sub-graph convolutional neural network layer until obtaining the spatial correlation features output by the last sub-graph convolutional neural network layer. Among them, the spatial correlation features output by the last sub-graph convolutional neural network layer are the spatial correlation features corresponding to the node feature sample data set: = A+ ; ; ; ; where represents the intermediate node-edge weight matrix, represents the Laplace equation matrix coefficient, represents node identification matrices (i.e., the identification matrices of ground deformation points), represents the new node-edge weight matrix; represents 's degree matrix, which is used for standard normalization during graph convolution operations; and are the spatial correlation features output by the -th layer sub-graph convolutional neural network and the -th layer sub-graph convolutional neural network in the graph convolutional neural network respectively, and are the initial activation function and the initial weight value of the graph convolutional neural network, respectively, and are the activation function and the weight value of the sub-graph convolutional neural network in the -th layer of the graph convolutional neural network, respectively, and are the activation function and the weight value of the sub-graph convolutional neural network in the -th layer of the graph convolutional neural network, respectively. It should be noted that the initial activation function of the graph convolutional neural network, the activation function of the sub-graph convolutional neural network in the -th layer of the graph convolutional neural network, and the activation function of the sub-graph convolutional neural network in the -th layer of the graph convolutional neural network are all ReLU (Rectified Linear Unit).

[0034] In an implementation manner of this embodiment, when inputting the spatial correlation features corresponding to the node feature sample data set into the initial hybrid ensemble learning network, the initial hybrid ensemble learning network performs slope stability evaluation on the spatial correlation features to obtain the slope stability evaluation result corresponding to the node feature sample data set, which specifically includes: Inputting the spatial correlation features corresponding to the node feature sample data set into the sub-network model layer, and multiple basic neural network models in the sub-network model layer perform preliminary slope stability evaluation based on the spatial correlation features and output multiple preliminary slope stability evaluation results; Inputting multiple preliminary slope stability evaluation results into the meta-machine learning classifier, and the meta-machine learning classifier performs logistic regression processing based on the multiple preliminary slope stability evaluation results and outputs the safety factor value of the InSAR ground deformation point corresponding to the node feature sample data set; When the safety factor value of the InSAR ground deformation point is greater than the preset threshold (in this embodiment, the preset threshold is taken as 1), it is determined that the slope stability evaluation result of the InSAR ground deformation point is stable; When the safety factor value of the InSAR ground deformation point is not greater than the preset threshold (in this embodiment, the preset threshold is taken as 1), it is determined that the slope stability evaluation result of the InSAR ground deformation point is unstable.

[0035] Specifically, as Figure 3As shown in the figure, the attention-based graph convolutional ensemble network AGCN-HEN mainly consists of two parts: AGCN (attention-graph convolutional network, an attention-based graph convolutional neural network) and HEN (hybrid ensemble learning network, a hybrid ensemble learning network). Among them, the AGCN network mainly embeds the attention mechanism Attention into the graph convolutional neural network GCN (graph convolutional network) to extract the spatial correlation features between ground deformation points; the HEN network includes a sub-network model at the first level and a meta-machine learning classifier at the second level to realize slope stability evaluation and output the safety factor value (i.e., the slope stability evaluation result).

[0036] As Figure 4 shown, the specific process of extracting the spatial correlation features between ground deformation points by the attention-based graph convolutional neural network AGCN includes: Let the node feature sample data set of the GGDN undirected network graph be , including input features of nodes, specifically including the deformation rates corresponding to each of the nodes and the slope stability influence factor values of nodes arranged in descending order based on the importance index. After the node feature sample data set extracted from the nodes of the GGDN undirected network graph enters the attention mechanism Attention layer, according to the ground deformation rate of each node in the graph and

[0037] the slope stability influence factor values of nodes arranged in descending order based on the importance index, the activation function in the Attention layer respectively obtains the weight values of nodes; and based on the weight value of each node and the slope stability input feature vector data (i.e., input feature) of this node, an attention matrix is obtained; then based on the attention matrix and the node edge weight matrix, the attention mechanism result (i.e., the attention feature set ) is obtained as the input data of the graph convolutional neural network GCN layer.

[0037] Taking the attention vector features (i.e., the attention feature set) obtained by the attention mechanism and the node edge weight matrix (i.e., the adjacency matrix) as the input data of the GCN network, the ground deformation spatial correlation features of all nodes of the GGDN undirected network graph are obtained through the GCN network. By constructing a stacked GCN network structure, the ground deformation spatial correlation feature data output by the previous layer GCN network structure of each node is used as the input data of the next layer GCN network structure, and so on, until the spatial correlation features between ground deformation points in the global area are obtained, and the spatial correlation features between ground deformation points are extracted through this GCN network structure.

[0038] As Figure 3 and Figure 5 shown, the specific process of the hybrid integrated learning network HEN for slope stability prediction (i.e., slope stability assessment) includes: The HEN network model mainly consists of two parts, including the sub-network model of the first layer and the meta-machine learning classifier of the second layer. The sub-network model of the first layer of the HEN network model contains 11 basic neural network models: LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), SRU (Simple Recurrent Unit), RF (Random Forest), AdaBoost (Adaptive Boosting), GBM (Gradient Boosting Machine), XGB (Extreme Gradient Boosting), SVM (Support Vector Machine), KNN (K-Nearest Neighbors), DT (Decision Tree), GNB (Gaussian Naive Bayes). The prediction results of these basic neural network models serve as the input data for the meta-machine learning classifier in the second layer of the HEN model. The HEN network takes the above spatially correlated features extracted by the AGCN network as a new sample data set (i.e., new input data), sets the network model parameters using the grid search algorithm, and each sub-network model in the HEN network model outputs the slope stability prediction result with the minimum error through continuous iterative training. Among them, LR represents Logistic Regression, represents the 1st new input data, represents the 2nd new input data, represents the 3rd new input data, represents the nth new input data, L represents the number of prediction objects (i.e., the number of InSAR ground deformation points), and P represents the number of sub-network models.

[0039] In an implementation of this embodiment, the error index of the model evaluation accuracy is used to describe the difference between the model prediction value (slope stability evaluation result) and the true value (slope stability calculation result), and can quantify the performance effect of the model. The present invention uses three statistical indices, namely the mean absolute error MAE (mean absolute error), the mean square error MSE (mean square error), and the root mean square error RMSE (Root Mean Square Error), to evaluate the evaluation performance effect of the slope stability evaluation model (i.e., the graph convolutional integration network based on the attention mechanism). The smaller the three indices are, the better the model performance effect is. The corresponding calculation formulas are as follows: ; ; ; Among them, represents the number of the test set, and respectively represent the th true value and the th predicted value. It should be noted that the present invention divides the node feature sample data set into a training set and a test set. In addition, slope stability prediction refers to slope stability evaluation, and the prediction result is the evaluation result.

[0040] The present invention proposes a graph convolutional neural network AGCN model embedded with an attention mechanism to extract the spatial correlation features between ground deformation points, and solves the problem that the existing slope stability evaluation methods do not consider the potential landslide hazards in the adjacent areas of landslides. At the same time, aiming at the problem that a single neural network is suitable for the analysis of homogeneous data, for a landslide data set with multi-source heterogeneity, by integrating the advantages of multiple single networks, a HEN hybrid integration network model is proposed to realize the application evaluation of multi-source heterogeneous structure data.

[0041] S3. Obtain the InSAR ground deformation point data to be evaluated, input the InSAR ground deformation point data to be evaluated into the graph convolutional integration network based on the attention mechanism, and output the slope stability evaluation result corresponding to the InSAR ground deformation point data to be evaluated.

[0042] In another implementation of this embodiment, the overall idea framework of the slope stability evaluation method of the graph convolutional integration network based on the attention mechanism is as Figure 2As shown in the figure, it is mainly divided into three parts: 1. Data preprocessing: First, through InSAR ground deformation point data and landslide point data, obtain InSAR ground deformation points related to landslide attributes, and extract multi-source slope stability influence factor attribute data based on this, and conduct importance analysis to obtain a slope stability influence sequence factor data set based on InSAR points. On this basis, construct global regional ground deformation spatial correlation data and the corresponding undirected network graph GGDN to provide data support for slope stability assessment. 2. AGCN-HEN network model: First, according to the proposed AGCN network, extract the spatial correlation characteristics between InSAR ground deformation points, and on this basis, use the proposed hybrid ensemble learning network HEN model to output the slope stability prediction result with the minimum error. 3. Accuracy evaluation: Use three error indices, namely mean absolute error MAE, mean square error MSE, and root mean square error RMSE, to evaluate the prediction accuracy of the AGCN-HEN network model.

[0043] In addition, based on the above slope stability assessment method of the graph convolutional ensemble network based on the attention mechanism, the present invention also provides a slope stability assessment system of the graph convolutional ensemble network based on the attention mechanism. Among them, a preferred embodiment of the slope stability assessment system of the graph convolutional ensemble network based on the attention mechanism is as Figure 6 shown, and specifically includes: Sample data set construction module 01: Used to obtain initial InSAR ground deformation point data and landslide point data, and preprocess the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the adjacent area of the landslide point and a node feature sample data set based on the InSAR ground deformation points; Network construction module 02: Used to construct an initial graph convolutional ensemble network based on the attention mechanism, train and test the initial graph convolutional ensemble network based on the attention mechanism using the node feature sample data set, and obtain a graph convolutional ensemble network based on the attention mechanism whose error index meets the evaluation accuracy requirements; Slope stability assessment module 03: Used to obtain the InSAR ground deformation point data to be evaluated, input the InSAR ground deformation point data to be evaluated into the graph convolutional ensemble network based on the attention mechanism, and output the slope stability assessment result corresponding to the InSAR ground deformation point data to be evaluated.

[0044] In addition, based on the above slope stability assessment method and system of the graph convolutional ensemble network based on the attention mechanism, the present invention also correspondingly provides a terminal. Among them, a preferred embodiment of the terminal is as Figure 7 shown, and specifically includes a processor 10, a memory 20, and a display 30. Figure 7Only some components of the terminal are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively.

[0045] The memory 20 can be an internal storage unit of the terminal in some embodiments, such as the hard disk or memory of the terminal. The memory 20 can also be an external storage device of the terminal in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, and a Flash Card equipped on the terminal, etc. Further, the memory 20 can also include both the internal storage unit and the external storage device of the terminal. The memory 20 is used to store application software installed on the terminal and various types of data, such as storing the program code of the terminal, etc. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, a slope stability evaluation program 40 of a graph convolutional integration network based on an attention mechanism is stored on the memory 20, and the slope stability evaluation program 40 of the graph convolutional integration network based on an attention mechanism can be executed by the processor 10, so as to implement the steps of the slope stability evaluation method of the graph convolutional integration network based on an attention mechanism in the present application.

[0046] The processor 10 can be a Central Processing Unit (CPU), a microprocessor, or other data processing chips in some embodiments, and is used to run the program code stored in the memory 20 or process data, such as executing the slope stability evaluation program 40 of the graph convolutional integration network based on an attention mechanism, etc.

[0047] The display 30 can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. in some embodiments. The display 30 is used to display information on the terminal and to display a visual user interface.

[0048] In one embodiment, when the processor 10 executes the slope stability evaluation program 40 of the graph convolutional integration network based on an attention mechanism in the memory 20, the steps of the slope stability evaluation method of the graph convolutional integration network based on an attention mechanism as described above are implemented.

[0049] The present invention also correspondingly provides a computer-readable storage medium, wherein the computer-readable storage medium stores a slope stability evaluation program of a graph convolutional integration network based on an attention mechanism, and when the slope stability evaluation program of the graph convolutional integration network based on an attention mechanism is executed by a processor, the steps of the slope stability evaluation method of the graph convolutional integration network based on an attention mechanism as described above are implemented.

[0050] It should be noted that, in this text, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or terminal. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or terminal including such element.

[0051] Of course, those of ordinary skill in the art can understand that all or part of the processes of implementing the above-described embodiment methods can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium readable by a computer. When the program is executed, it can include the processes of the above-described method embodiments. The computer-readable storage medium can be a memory, a magnetic disk, an optical disc, etc.

[0052] It should be understood that the application of the present invention is not limited to the above examples. For those of ordinary skill in the art, improvements or transformations can be made according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention.

Claims

1. A slope stability assessment method based on a graph convolutional ensemble network with an attention mechanism, characterized in that: The slope stability assessment method of the graph convolutional integrated network based on the attention mechanism includes: Acquiring initial InSAR ground deformation point data and landslide point data, and preprocessing the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the vicinity of the landslide point and a node feature sample data set based on the InSAR ground deformation points; Constructing an initial graph convolutional integration network based on the attention mechanism, training and testing the initial graph convolutional integration network based on the attention mechanism using the node feature sample data set, and obtaining a graph convolutional integration network based on the attention mechanism whose error index meets the evaluation accuracy requirements; The InSAR ground deformation point data to be evaluated is obtained, the InSAR ground deformation point data to be evaluated is input into the graph convolutional integration network based on the attention mechanism, and the slope stability evaluation result corresponding to the InSAR ground deformation point data to be evaluated is output.

2. The slope stability assessment method based on graph convolutional integrated network with attention mechanism according to claim 1 is characterized in that: The step of acquiring initial InSAR ground deformation point data and landslide point data, and preprocessing the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the vicinity of the landslide point and a node feature sample data set based on the InSAR ground deformation points specifically includes: Obtain the initial InSAR ground deformation point data, landslide point data, public multi-source landslide inducing factor data, ground elevation data and slope data of the study area; According to the initial InSAR ground deformation point data, the landslide point data, the disclosed multi-source landslide inducing factor data, the ground elevation data and the slope data, the InSAR ground deformation point with the associated slope stability influencing factor sequence and safety factor value attribute in the area adjacent to the landslide point is obtained through spatial correlation analysis, spatial multi-value extraction analysis, importance sorting and safety factor value calculation; Based on the InSAR ground deformation points associated with the slope stability influencing factor sequence and the safety factor value attribute, an undirected network graph is constructed through spatial correlation calculation, and a node feature sample data set based on the InSAR ground deformation points is obtained using the undirected network graph.

3. The slope stability assessment method based on graph convolutional integrated network with attention mechanism according to claim 2 is characterized in that: The method of obtaining the InSAR ground deformation point data, the landslide point data, the disclosed multi-source landslide inducing factor data, the ground elevation data and the slope data through spatial correlation analysis, spatial multi-value extraction analysis, importance sorting and safety factor value calculation of the associated slope stability influencing factor sequence and safety factor value attribute of the area adjacent to the landslide point specifically includes: According to the initial InSAR ground deformation point data and the landslide point data, establishing attribute association between the landslide point and the initial InSAR ground deformation point through spatial association analysis, and obtaining InSAR ground deformation points associated with landslide attributes, wherein the InSAR ground deformation points are initial InSAR ground deformation points in the vicinity of the landslide point; According to the InSAR ground deformation points and the disclosed multi-source landslide inducing factor data, multiple slope stability influencing factors are obtained through spatial multi-value extraction analysis; An information gain algorithm is used to analyze the importance index of each of the slope stability influencing factors, and each of the slope stability influencing factors is arranged in descending order according to the importance index of each of the slope stability influencing factors to obtain a slope stability influencing factor sequence; Calculating the safety factor value of the InSAR ground deformation point using a finite element method according to the ground elevation data and the slope data corresponding to the InSAR ground deformation point; Based on the InSAR ground deformation point, the slope stability influencing factor sequence and the safety factor value, an InSAR ground deformation point with associated slope stability influencing factor sequence and safety factor value attributes in an area adjacent to the landslide point is obtained; The InSAR ground deformation point data corresponding to the InSAR ground deformation point associated with the slope stability influencing factor sequence and the safety factor value attribute include the safety factor value, the deformation rate, the slope stability influencing factor sequence and the safety factor value.

4. The slope stability assessment method based on the graph convolutional integrated network with attention mechanism according to claim 3 is characterized in that: The InSAR ground deformation points based on the associated slope stability influencing factor sequence and safety factor value attribute are constructed by spatial correlation calculation, and the node feature sample data set based on the InSAR ground deformation points is obtained by using the undirected network graph, specifically including: Based on the InSAR ground deformation points associated with the slope stability influencing factor sequence and the safety factor value attribute, spatial correlation data based on the InSAR ground deformation points are obtained through spatial correlation calculation, wherein the spatial correlation data includes a spatial correlation value, and the spatial correlation value is calculated by a Gaussian similarity function: =exp ; in, Indicates InSAR ground deformation point and The spatial correlation value between InSAR ground deformation points, Indicates InSAR ground deformation point and The Euclidean distance between InSAR ground deformation points, Indicates InSAR ground deformation point and The Euclidean distance between InSAR ground deformation points The standard deviation of , exp represents the exponential function; According to the InSAR ground deformation points associated with the slope stability influencing factor sequence and the safety factor value attribute and the spatial correlation data based on the InSAR ground deformation points, an undirected network graph is constructed, the InSAR ground deformation points are used as nodes of the undirected network graph, the deformation rates and slope stability influencing factor sequences of the InSAR ground deformation points are used as the deformation rates and slope stability influencing factor sequences of the corresponding nodes, and the spatial correlation values ​​between the InSAR ground deformation points are used as the spatial similarity values ​​and node edge weights between the corresponding nodes to obtain the undirected network graph. , where v represents the undirected network graph g A collection of nodes, represents the edge data set, A represents the node edge weight matrix; The node feature sample data set is obtained by using the undirected network graph, wherein the node feature sample data set includes The corresponding nodes Input features include deformation rate and slope stability influencing factor sequence: ； ； in, represents the node feature sample dataset, represents the input feature corresponding to the first node, Represents the input feature corresponding to the second node, Indicates The input features corresponding to the nodes are Indicates The input features corresponding to the nodes are Represents the total number of nodes, Indicates The deformation rate corresponding to each node is Indicates The first slope stability influencing factor corresponding to the node is: Indicates The second slope stability influencing factor corresponding to the node is: Indicates The node corresponding to The factors affecting slope stability are: Represents the total number of slope stability influencing factors in the slope stability influencing factor sequence.

5. The slope stability assessment method based on graph convolutional integrated network with attention mechanism according to claim 4 is characterized in that: The initial graph convolutional integration network based on the attention mechanism is constructed, and the initial graph convolutional integration network based on the attention mechanism is trained and tested using the node feature sample data set to obtain a graph convolutional integration network based on the attention mechanism whose error index meets the evaluation accuracy requirement, specifically including: Constructing an initial graph convolutional ensemble network based on an attention mechanism, wherein the initial graph convolutional ensemble network based on an attention mechanism includes an initial graph convolutional neural network based on an attention mechanism and an initial hybrid ensemble learning network, the initial graph convolutional neural network based on an attention mechanism includes an attention mechanism layer and a graph convolutional neural network, and the initial hybrid ensemble learning network includes a subnetwork model layer and a meta-machine learning classifier; Inputting the node feature sample data set into the initial graph convolutional neural network based on the attention mechanism, the initial graph convolutional neural network based on the attention mechanism performs spatial correlation feature extraction on the node feature sample data set to obtain the spatial correlation feature corresponding to the node feature sample data set; The spatial correlation features corresponding to the node feature sample data set are input into the initial hybrid ensemble learning network, and the initial hybrid ensemble learning network performs slope stability assessment on the spatial correlation features to obtain the slope stability assessment result corresponding to the node feature sample data set, wherein the slope stability assessment result corresponding to the node feature sample data set is the slope stability assessment result corresponding to the InSAR ground deformation point; Comparing the slope stability assessment result corresponding to the InSAR ground deformation point with the slope stability calculation result corresponding to the safety factor value in the InSAR ground deformation point data, iteratively training and testing the initial graph convolutional integration network based on the attention mechanism according to the comparison result, until the error index of the initial graph convolutional integration network based on the attention mechanism meets the assessment accuracy requirement, thereby obtaining the graph convolutional integration network based on the attention mechanism; The error index includes mean absolute error, mean square error and root mean square error.

6. The slope stability assessment method based on the graph convolutional integrated network with attention mechanism according to claim 5 is characterized in that: The step of inputting the node feature sample data set into the initial graph convolutional neural network based on the attention mechanism, and extracting spatial correlation features from the node feature sample data set to obtain spatial correlation features corresponding to the node feature sample data set, specifically includes: The node feature sample data set is input into the attention mechanism layer. The activation function of the attention mechanism layer calculates the weight value of each node according to the deformation rate and slope stability influencing factor sequence of each node in the node feature sample data set, and calculates the attention matrix based on the weight value of each node and the input feature: ； in, represents the attention matrix, represents the activation function of the attention mechanism layer, Indicates The weight value of the node, Indicates The deviation of each node; The attention matrix is ​​normalized to obtain a normalized attention matrix: ； in, Represents the normalized attention matrix 'The weight, Represents the attention matrix The amount of Multiply the normalized attention matrix by the node edge weight matrix to obtain the attention feature set ; The attention feature set Input into the graph convolutional neural network, each layer of subgraph convolutional neural network in the graph convolutional neural network outputs spatial correlation features based on the attention feature set and the node edge weight matrix, and uses the spatial correlation features as the input data of the next layer of subgraph convolutional neural network until the spatial correlation features output by the last layer of subgraph convolutional neural network are obtained, wherein the spatial correlation features output by the last layer of subgraph convolutional neural network are the spatial correlation features corresponding to the node feature sample data set: = A+ ； ； ； ； in, represents the intermediate node edge weight matrix, represents the coefficients of the Laplace equation matrix, express The identification matrix of nodes, represents the new node edge weight matrix, express The degree matrix of and They are respectively Layer subgraph convolutional neural network and the The spatial correlation characteristics of the output of the layer subgraph convolutional neural network, and are the initial activation function and initial weight value of the graph convolutional neural network, and They are respectively The activation function and weight value of the layer subgraph convolutional neural network, and They are respectively Activation functions and weight values ​​of the layer subgraph convolutional neural network.

7. The slope stability assessment method based on graph convolutional integrated network with attention mechanism according to claim 6 is characterized in that: The step of inputting the spatial correlation features corresponding to the node feature sample data set into the initial hybrid ensemble learning network, wherein the initial hybrid ensemble learning network performs slope stability assessment on the spatial correlation features to obtain the slope stability assessment result corresponding to the node feature sample data set, specifically includes: Inputting the spatial correlation features corresponding to the node feature sample data set into the sub-network model layer, multiple basic neural network models in the sub-network model layer perform preliminary slope stability assessment based on the spatial correlation features, and outputting multiple preliminary slope stability assessment results; Inputting the multiple preliminary slope stability assessment results into the meta-machine learning classifier, the meta-machine learning classifier performs logistic regression processing based on the multiple preliminary slope stability assessment results, and outputs the safety factor value of the InSAR ground deformation point corresponding to the node feature sample data set; When the safety factor value of the InSAR ground deformation point is greater than a preset threshold, determining that the slope stability assessment result of the InSAR ground deformation point is stable; When the safety factor value of the InSAR ground deformation point is not greater than a preset threshold, the slope stability assessment result of the InSAR ground deformation point is determined to be unstable.

8. A slope stability assessment system based on a graph convolutional ensemble network with an attention mechanism, characterized in that: The slope stability assessment system based on the graph convolutional integrated network of the attention mechanism includes: Sample data set construction module: used to obtain initial InSAR ground deformation point data and landslide point data, and pre-process the initial InSAR ground deformation point data and the landslide point data to obtain InSAR ground deformation points in the vicinity of the landslide point and node feature sample data sets based on the InSAR ground deformation points; Network construction module: used to construct an initial graph convolutional integration network based on the attention mechanism, train and test the initial graph convolutional integration network based on the attention mechanism using the node feature sample data set, and obtain a graph convolutional integration network based on the attention mechanism whose error index meets the evaluation accuracy requirements; Slope stability assessment module: used to obtain the InSAR ground deformation point data to be assessed, input the InSAR ground deformation point data to be assessed into the graph convolutional integration network based on the attention mechanism, and output the slope stability assessment result corresponding to the InSAR ground deformation point data to be assessed.

9. A terminal, characterized in that: The terminal includes: a memory, a processor, and a slope stability assessment program of a graph convolutional integrated network based on an attention mechanism, which is stored in the memory and can be run on the processor. When the slope stability assessment program of a graph convolutional integrated network based on an attention mechanism is executed by the processor, the steps of the slope stability assessment method of a graph convolutional integrated network based on an attention mechanism are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a slope stability assessment program of a graph convolutional integration network based on an attention mechanism. When the slope stability assessment program of a graph convolutional integration network based on an attention mechanism is executed by a processor, the steps of the slope stability assessment method of a graph convolutional integration network based on an attention mechanism as described in any one of claims 1 to 7 are implemented.

Images