Complex high-position landslide risk dynamic identification system based on multi-source InSAR cooperation

Abstract

The invention relates to the technical field of geological disaster monitoring and early warning, in particular to a complex high-position landslide risk dynamic identification system based on multi-source InSAR cooperation, and the system comprises a data collection module which is used for constructing a normalized feature vector representing the state of each monitoring node; the correlation modeling module is used for quantifying the physical influence degree among the heterogeneous nodes; the state prediction module is used for iteratively updating the hidden state of each heterogeneous node by adopting a space-time diagram neural network model and interpreting the hidden state into incremental displacement; the risk assessment module is used for calculating a risk precursor index of each heterogeneous node deviating from a normal evolution mode based on the hidden state, and judging and issuing graded early warning according to a preset threshold system; according to the method, the overfitting problem of a pure data driving model during data sparsity is effectively relieved, and the generalization ability and the physical interpretability of a prediction result are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological disaster monitoring and early warning, in particular to a complex high-position landslide risk dynamic identification system based on multi-source InSAR cooperation. BACKGROUND

[0002] With the deepening of geological disaster prevention work, complex high-position landslides are strong in concealment and severe in disaster, and accurate dynamic risk identification thereof becomes the key to protecting people's lives and property. However, landslide is a complex geological body with spatio-temporal evolution, and its risk identification faces great challenges. Traditional landslide risk identification methods mainly rely on single-source monitoring data or fuse multi-source data after interpretation, which not only leads to loss of source information and time delay, but also is difficult to fully depict the whole process of landslide incubation and evolution. The existing technology has the following limitations: limitation of data fusion, traditional monitoring methods usually fuse after calculating physical quantities such as displacement, which breaks the internal correlation between data and causes information loss. Meanwhile, different sensors have different modalities, frequencies and dimensions, making it difficult to effectively fuse features at the original level; limitation of correlation modeling, existing models often ignore the complex physical mechanisms within the landslide body, such as only considering spatial distance when building node correlation without fully incorporating key geological prior knowledge such as elevation difference and gravity effect, resulting in a model that cannot truly reflect the stress and displacement transmission path; limitation of state prediction, data-driven prediction models are prone to overfitting when monitoring data is sparse or noisy, and their prediction results may violate basic physical laws, lacking generalization ability and interpretability; Temporal graph neural network technology has great potential in processing data with complex spatial topology and temporal evolution characteristics. Deep fusion of multi-source monitoring data at the feature level and embedding of physical constraints for model training can significantly improve the accuracy and reliability of prediction. In summary, the existing technology has limitations in deep fusion of multi-source heterogeneous data, spatio-temporal correlation modeling with geological prior knowledge, and data and physical dual-driven prediction, making it difficult to accurately and dynamically identify and warn complex high-position landslides. SUMMARY

[0003] To solve the above technical problems, the present application provides a complex high-position landslide risk dynamic identification system based on multi-source InSAR cooperation. Specifically, the technical solution of the present application includes: A data acquisition module for acquiring multi-source heterogeneous sensor data streams of a landslide monitoring area and constructing normalized feature vectors representing the state of each monitoring node based on the multi-source heterogeneous sensor data streams. The correlation modeling module is configured to discretize the landslide monitoring area into heterogeneous nodes and quantify the degree of physical influence between the heterogeneous nodes based on a weighted adjacency matrix containing geological prior knowledge. The state prediction module is configured to iteratively update the hidden state of each heterogeneous node based on the normalized feature vector and the weighted adjacency matrix using a spatio-temporal graph neural network model, and interpret the hidden state as an incremental displacement. The risk assessment module is configured to calculate a risk precursor index of each heterogeneous node deviating from the normal evolution mode based on the hidden state, and determine and issue a graded early warning according to a preset threshold system.

[0004] Preferably, the multi-source heterogeneous sensor data stream includes raw interferometric phase stream data and coherence stream data, which are obtained by an InSAR satellite; the multi-source heterogeneous sensor data stream further includes a three-dimensional real-time displacement vector stream, which is obtained by a GNSS station; and the multi-source heterogeneous sensor data stream further includes pulse rainfall data, which is obtained by a rain gauge.

[0005] Preferably, the correlation modeling module is configured to combine a Gaussian attenuation effect based on spatial distance and a terrain influence effect based on elevation difference to determine the weighted adjacency matrix. The terrain influence effect simulates the influence asymmetry under the action of gravity through a nonlinear transformation to enhance the influence of upstream nodes on downstream nodes.

[0006] Preferably, the state prediction module iteratively updates the hidden state of each heterogeneous node, including: performing weighted aggregation on the hidden state of each neighbor node at the previous time through a graph convolution operation to obtain a comprehensive influence representing spatial neighborhood transmission; feeding the comprehensive influence and the normalized feature vector of the heterogeneous node at the current time into a gated recurrent unit to calculate the hidden state at the current time.

[0007] Preferably, the spatio-temporal graph neural network model is trained using a hybrid loss function; the hybrid loss function includes a prediction accuracy loss and a physical regularization term.

[0008] Preferably, the physical regularization term is used to impose a physical constraint during model training. The physical constraint is that the deformation rate of the landslide should not slow down when effective rainfall occurs.

[0009] Preferably, the risk assessment module calculates the risk precursor index, including: using stable historical monitoring data to generate a set of hidden state vectors representing a normal state through the trained spatio-temporal graph neural network model; based on the hidden state vectors representing the normal state, calculating the average hidden state and standard deviation of each node; The risk precursor index is obtained by measuring the Euclidean distance between the hidden state at the current moment and the average hidden state, and normalizing it with the standard deviation.

[0010] Preferably, the hierarchical early warning includes: When the risk precursor index is less than or equal to the first preset threshold, it is determined to be a normal state and no early warning is issued; When the risk precursor index is greater than the first preset threshold and less than or equal to the second preset threshold, a concern level early warning is issued; When the risk precursor index is greater than the second preset threshold, a risk level early warning is issued.

[0011] Compared with the prior art, the present application has the following beneficial effects: 1. The system fuses features at the data source, and constructs a unified normalized feature vector from InSAR, GNSS, rainfall and other multi-source heterogeneous data streams in real time, which avoids the information loss and time delay caused by traditional methods of solving and then fusing, provides high-quality input for accurate prediction, and improves the real-time performance of risk identification; 2. The system builds a weighted adjacency matrix containing geological prior knowledge, which not only considers the spatial distance between nodes, but also innovatively incorporates the topographic influence effect caused by elevation difference. This method can simulate the asymmetric influence under gravity, making the model more realistically reflect the transmission path of stress and displacement within the landslide body, and improving the physical authenticity of the model; 3. The system adopts a data and physical dual-driven training paradigm, and introduces a physical regularization term in model training to impose constraints on physical laws such as the fact that the landslide deformation rate should not slow down during effective rainfall, which effectively alleviates the overfitting problem of pure data-driven models when data is sparse, significantly improving the generalization ability and physical interpretability of the prediction results; 4. The risk assessment of the system no longer relies on a single displacement threshold, but calculates the risk precursor index by measuring the deviation of the high-dimensional hidden state within the model from the normal benchmark. Since the hidden state incorporates temporal and spatial information, it can more comprehensively represent the node's overall state, thereby enabling earlier and more sensitive capture of instability precursors and improving the sensitivity and forward-looking nature of the early warning. BRIEF DESCRIPTION OF DRAWINGS

[0012] The present application will be further explained in conjunction with the accompanying drawings and examples: Figure 1 is a structural diagram of the system of the present application. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with specific examples.

[0014] Example 1: Referring to Figure 1 , a complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation comprises: A data acquisition module is configured to acquire multi-source heterogeneous sensor data streams of a landslide monitoring area, and construct normalized feature vectors representing the states of each monitoring node based on the multi-source heterogeneous sensor data streams; An association modeling module is configured to discretize the landslide monitoring area into heterogeneous nodes, and quantify the physical influence degree between the heterogeneous nodes based on a weighted adjacency matrix containing geological prior knowledge; A state prediction module is configured to iteratively update the hidden states of each heterogeneous node based on the normalized feature vectors and the weighted adjacency matrix using a spatio-temporal graph neural network model, and interpret the hidden states as incremental displacements; A risk assessment module is configured to calculate the risk precursor index of each heterogeneous node deviating from the normal evolution mode based on the hidden states, and determine and issue a graded early warning according to a preset threshold system.

[0015] The embodiment of the present application provides a complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation, which aims to realize accurate and dynamic identification and early warning of complex high-order landslide risk through multi-source data fusion, spatio-temporal association modeling and prediction of physical mechanism embedding; the system comprises a data acquisition module, an association modeling module, a state prediction module and a risk assessment module; The purpose of the data acquisition module is to acquire multi-source heterogeneous sensor data of the monitoring area in real time and asynchronously, and convert it into a unified and standardized data structure, providing high-quality input for subsequent cross-modal analysis and prediction; in this embodiment, the module acquires multi-source heterogeneous sensor data streams of the landslide monitoring area through corresponding sensors; to solve the differences in modal, frequency and dimension of different data streams, the module constructs normalized feature vectors representing the states of each monitoring node based on these data streams; the implementation manner is that at any time point , the original or semi-processed data of each sensor is mapped to the normalized feature vector of the corresponding node in the monitoring graph through a preset transformation function

[0016] , wherein : the normalized feature vector of the i th monitoring node in the graph at time t is calculated and generated by the data acquisition module as the real-time and multi-modal input of the subsequent spatio-temporal graph neural network model; : the normalized feature vector of the i th monitoring node in the graph at time t is calculated and generated by the data acquisition module as the real-time and multi-modal input of the subsequent spatio-temporal graph neural network model; : the normalized feature vector of the i th monitoring node in the graph at time t is calculated and generated by the data acquisition module as the real-time and multi-modal input of the subsequent spatio-temporal graph neural network model; ​The original interference phase value of the node at the moment, the data type is floating point number, and the source is collected by the InSAR satellite sensor; : The coherence value corresponding to the phase value, the data type is floating point number, and the source is collected by the InSAR satellite sensor; : The three-dimensional real-time displacement vector measured by the GNSS site, the data type is a three-dimensional vector, and the source is collected by the ground GNSS site; for the nodes without GNSS coverage, the parameter can be estimated by spatial interpolation methods such as Kriging interpolation or inverse distance weighting interpolation according to the data of adjacent GNSS sites to ensure the uniformity of the dimension of the feature vector; : The regional rainfall affecting the first node, the data type is floating point number, and the source is collected by the ground rain gauge; : The preset normalization function for various types of heterogeneous data, which maps the original data of different dimensions to similar numerical intervals to eliminate dimensional differences and ensure the convergence and stability of model training, and the source is determined in advance according to the statistical analysis of historical data of the monitoring area; for example, the minimum-maximum normalization method can be used to convert any original data to , where and are the minimum and maximum values of the corresponding physical quantity obtained according to historical data statistics; The purpose of the correlation modeling module is to organize discrete monitoring nodes into a network structure that can reflect the geological and physical characteristics of the landslide, and to provide a structured path for the effective propagation of spatio-temporal information; in this embodiment, the module discretizes the entire landslide monitoring area into a dynamic graph composed of heterogeneous nodes, where is the node set, and is the edge set; the nodes are defined according to the data source, such as InSAR pixel nodes, GNSS site nodes, etc.; this module quantifies the physical influence degree between heterogeneous nodes based on the weighted adjacency matrix containing geological prior knowledge; the value of the element in the matrix represents the physical influence degree of node on node ; The purpose of the state prediction module is to fuse the characteristics of the nodes and the influence of the neighborhood to predict the future state evolution of each monitoring point in the landslide; in this embodiment, the module is based on the normalized feature vector A spatiotemporal graph neural network model is used to iteratively update the hidden states of each heterogeneous node. Hidden state It is a high-dimensional vector for monitoring nodes. exist The abstract representation of the physical state at any given moment contains a comprehensive result of historical information, current input, and the influence of spatial neighborhood. This module interprets this high-dimensional, abstract hidden state into incremental displacements with clear physical meaning through a fully connected output layer. ; The purpose of the risk assessment module is to perform real-time analysis of the internal states predicted by the model, dynamically quantify the risk of landslides evolving from a stable state to an unstable state, and provide graded early warnings. In this embodiment, this module is based on the hidden states output by the state prediction module. By measuring the degree of deviation from the normal state distribution, the risk precursor index of each heterogeneous node from the normal evolutionary pattern is calculated. Risk precursor index It is a dimensionless scalar whose value directly reflects the degree of abnormality of the current state of the monitoring node; the module judges the calculated risk precursor index based on a preset threshold system with clear statistical significance and issues corresponding graded warnings. This embodiment constructs an end-to-end dynamic identification system, realizing a complete technical closed loop from multi-source heterogeneous data acquisition, physical correlation modeling, spatiotemporal state prediction to risk assessment and early warning. The system performs feature-level fusion directly at the raw data level, avoiding the information loss and time delay caused by the traditional solution-then-fusion paradigm. Through the graph structure containing geological a priori information, the model can capture the complex physical mechanisms within the landslide. Using a spatiotemporal graph neural network, it effectively integrates temporal evolution and spatial influence. Finally, through anomaly detection of the model's internal state, it achieves dynamic and quantitative assessment of landslide risk. This system significantly improves the real-time performance, accuracy, and interpretability of complex high-altitude landslide risk identification.

[0017] Example 2: The multi-source heterogeneous sensor data stream includes raw interferometric phase stream data and coherent stream data, which are acquired by InSAR satellites; the multi-source heterogeneous sensor data stream also includes three-dimensional real-time displacement vector stream, which is acquired by GNSS stations; the multi-source heterogeneous sensor data stream also includes pulsed rainfall data, which is acquired by rain gauges.

[0018] Based on the implementation method described in Example 1, this embodiment defines the specific composition of the multi-source heterogeneous sensor data stream in order to ensure the comprehensiveness and complementarity of the input data, so as to more accurately capture the deformation and inducing factors of the landslide body. The multi-source heterogeneous sensor data stream includes: raw interferometric phase stream data and coherence stream data acquired by an InSAR satellite; three-dimensional real-time displacement vector stream acquired by a GNSS station; and pulsed rainfall data acquired by a rain gauge; By explicitly combining the large-scale monitoring capability of InSAR, the high-precision fixed-point observation capability of GNSS, and the quantification capability of the rain gauge on the core inducing factor, the system obtains comprehensive observation of the landslide deformation-inducing chain; the synergy of the multi-source data enables the normalized feature vector to more completely depict the physical state of the monitoring node at any moment, thereby providing more abundant and reliable information input for the state prediction module, and directly improving the prediction accuracy and robustness of the subsequent risk identification model.

[0019] Embodiment 3: The correlation modeling module is configured to determine a weighted adjacency matrix by combining a Gaussian attenuation effect based on spatial distance and a topographic influence effect based on elevation difference. The topographic influence effect simulates the influence asymmetry under the action of gravity through a nonlinear transformation to enhance the influence of an upstream node on a downstream node.

[0020] On the basis of the implementation manner described in Embodiment 1, the present embodiment specifically limits how the correlation modeling module determines the weighted adjacency matrix , so that the quantification depth of the correlation between nodes is integrated with geological prior knowledge conforming to the landslide dynamics characteristics. The correlation modeling module calculates the adjacency matrix element through a comprehensive model that combines a Gaussian attenuation effect based on spatial distance and a topographic influence effect based on elevation difference. The internal logic of the topographic influence effect is to simulate the influence asymmetry under the action of gravity, thereby significantly enhancing the influence of an upstream node on a downstream node. The two effects are fused through the following formula to determine the element in the weighted adjacency matrix :

[0021] wherein, is a weight value quantifying the influence strength of a node on a node , and is a floating-point number calculated and generated by the correlation modeling module; is the spatial distance between a node and a node , and is calculated according to the geographical coordinates of the nodes. : This is the distance attenuation coefficient, a hyperparameter used to control the range of influence. Its setting is based on the characteristic scale of the landslide body in the geological survey report to ensure that the model conforms to the scale effect of the actual geological unit. : for nodes and elevation difference Its source is calculated based on the digital elevation model data of the nodes; : This is the elevation influence factor, and its dimension is the reciprocal of length, such as As a hyperparameter that adjusts the contribution of elevation difference to the asymmetry of influence weights, its value is optimized and determined by cross-validation of model performance on the validation set. This embodiment transcends the limitations of traditional graph models that only consider spatial distance. By asymmetrically embedding the gravity effect caused by elevation differences into the adjacency matrix, the constructed landslide physical correlation network topology more realistically reflects the transmission path of stress and displacement within the landslide, providing a more physically consistent information propagation framework for subsequent spatiotemporal graph neural networks and improving prediction accuracy.

[0022] Example 4: The state prediction module iteratively updates the hidden states of each heterogeneous node, including: By performing graph convolution operations, the hidden states of each neighboring node at the previous time step are weighted and aggregated to obtain the comprehensive influence representing the spatial neighborhood transmission. The combined influence and the normalized feature vector of the heterogeneous node at the current moment are fed into the gate control loop unit to calculate the hidden state at the current moment.

[0023] Based on the implementation method described in Example 1, this example describes how the state prediction module iteratively updates the hidden states of each heterogeneous node. It provides a specific implementation process that integrates the powerful spatial information aggregation capability of graph convolutional networks with the excellent time-series information processing capability of gated recurrent units; At each time step Through graph convolution operations, the target node is... Each neighbor node In the previous hidden state Perform weighted aggregation to obtain the representation space neighborhood and pass it to the node. The overall influence vector; The combined influence vector and nodes The normalized eigenvector at the current moment The GRU is jointly fed, selectively forgets historical information and integrates new information through its internal update gate and reset gate mechanisms, and finally calculates the node hidden state at the current time The entire iterative update process can be described by the following formula:

[0024] wherein, is the node The updated hidden state at the time is the high-dimensional abstract representation of the physical state of the node at the current time, which is calculated by the state prediction module; is a GRU function, which is used to integrate current input and historical information to update the state; is the node external input feature vector at the time , which is provided by the data acquisition module; is a graph convolution operation, which is used to aggregate historical state information from the neighborhood , wherein is provided by the association modeling module, is the hidden state at the previous time, is a weight matrix to be learned by the model during training, which is used to linearly transform the hidden state; By combining graph convolution with GRU, a spatio-temporal information processing unit is constructed; the graph convolution operation effectively captures the spatial interaction within the landslide body at each time step, and the GRU further processes the time dependence of the deformation process, thereby more accurately predicting the landslide dynamic behavior driven by spatial effects and time accumulation; at time step , the initial hidden state of each node can be initialized as a zero vector; To achieve the final displacement prediction, the state prediction module further maps the hidden state at the current time to the physically interpretable incremental displacement prediction value through a decoder such as a fully connected output layer; this process can be represented by the following formula: wherein represents a multi-layer perceptron or a single fully connected layer, the weights of which are learned during model training.

[0025] Embodiment 5: The spatio-temporal graph neural network model is trained using a hybrid loss function; the hybrid loss function includes a prediction accuracy loss and a physical regularization term; The physical regularization term is used to impose a physical constraint in the model training; wherein the physical constraint is that the deformation rate of the landslide should not slow down when effective rainfall occurs.

[0026] The training mechanism of the spatio-temporal graph neural network model is optimized in this embodiment, and a hybrid loss function is used for training, aiming to ensure that the prediction result of the model not only has high data fitting accuracy, but also meets the established physical law; The hybrid loss function is composed of a prediction accuracy loss and a physical regularization term; the prediction accuracy loss is a data-driven term, which is used to measure the gap between the model prediction value and the true observation value; the physical regularization term is a knowledge-driven term, which is used to impose a clear physical constraint in the model training process, and encode the prior physical knowledge into a differentiable penalty term; in this embodiment, the physical constraint imposed is that the deformation rate of the landslide should not slow down when effective rainfall occurs, and the complete hybrid loss function is which is defined by the following formula:

[0027] wherein, : is the total loss function, which is the optimization objective of the model training, defined in this embodiment; : is the total number of nodes participating in the calculation; : is the square of the Euclidean distance, which is used to calculate the prediction accuracy loss; wherein is the incremental displacement predicted by the model, is the true incremental displacement observation value; : is the reference displacement, which is used to non-dimensionalize the prediction accuracy loss term to ensure that the dimensions of the components of the hybrid loss function are consistent, and its source can be determined according to the displacement amplitude range of the historical monitoring data; : is the weight coefficient, which is a hyperparameter, used to balance the fitting accuracy of the model to the observation data and the degree of compliance with the physical law, and its value is determined by cross-validation; : is the rectified linear unit function, which is used here to realize one-way punishment: only when the physical constraint is violated, a penalty term is generated; : is the normalized rate , wherein is calculated from the predicted incremental displacement, and the reference rate is determined from historical data statistics or geological survey reports; : is the normalized rainfall wherein is the real-time rainfall, and the critical rainfall threshold The source is determined by historical data statistics or geological survey reports. By introducing a hybrid loss function containing a physical regularization term, prior geological knowledge is injected into the training process of the model. This data-physical dual-driven training paradigm effectively alleviates the overfitting problem that may occur when a purely data-driven model is used in sparse or noisy monitoring data, improving the model's generalization ability, while ensuring that the model's prediction results are physically reasonable and interpretable.

[0028] Embodiment 6: The risk assessment module calculates a risk precursor index, including: Using stable historical monitoring data, a set of hidden state vectors representing normal state is generated by the trained spatio-temporal graph neural network model. Based on the hidden state vectors representing the normal state, the average hidden state and the standard deviation of each node are calculated. By measuring the Euclidean distance between the current hidden state and the average hidden state, and normalizing it with the standard deviation, the risk precursor index is obtained.

[0029] Based on the implementation described in Embodiment 1, this embodiment describes how the risk assessment module calculates the risk precursor index A detailed implementation process is provided, the core of which is to identify risk precursors by quantifying the deviation of the model's internal state vector from the normal baseline. In the initial stage of system deployment, a large amount of historical monitoring data labeled as stable, for example, data in which the cumulative displacement and deformation rate of the node are below the pre-set geological safety threshold within a certain monitoring period, is input into the trained spatio-temporal graph neural network model to generate a set of hidden state vectors representing normal state. Based on this set of normal state vectors, the average hidden state and the standard deviation of each monitoring node are calculated. ; In the real-time monitoring stage, the Euclidean distance between the current hidden state and the historical average hidden state of the node is measured, and the standard deviation of the node's history is used to normalize this distance to obtain a dimensionless risk precursor index , which is defined by the following formula:

[0030] wherein, : risk precursor index, dynamic quantification of node In the degree of deviation from its normal evolution pattern at the current moment, calculated by the risk assessment module; : average hidden state of node at the current moment , which is output in real time by the state prediction module; : average hidden state of node , which is the reference center of the normal state of the node, and is obtained by statistically averaging the hidden states under a large amount of stable historical data; : standard deviation of the Euclidean norm of the hidden state vector of node , which is a normalized scalar scale to measure the statistical significance of deviation, and is obtained by statistically calculating the Euclidean norm of the hidden state vector under a large amount of stable historical data; : a very small positive number, which prevents the denominator from being zero due to the standard deviation being zero in the calculation; The embodiment deepens the basis of risk identification from a single physical output to the high-dimensional hidden state inside the model; Since the hidden state integrates temporal and spatial information and multi-source input, it can more comprehensively represent the comprehensive state of the node than any single physical quantity, thereby capturing weak abnormal signals in the quantitative to qualitative change process of the landslide earlier and more sensitively, and improving the sensitivity and foresight of the early warning.

[0031] Embodiment 7: The graded early warning includes: When the risk precursor index is less than or equal to the first preset threshold, it is determined to be in a normal state and no early warning is issued; When the risk precursor index is greater than the first preset threshold and less than or equal to the second preset threshold, a concern-level early warning is issued; When the risk precursor index is greater than the second preset threshold, a risk-level early warning is issued.

[0032] Based on the implementation manner described in embodiment 1, the embodiment specifically implements the determination logic of the graded early warning, and converts the continuously changing risk precursor index into discrete early warning levels with clear disposal guidance significance; The graded early warning system of the embodiment establishes a three-level response mechanism based on the first preset threshold and the second preset threshold: When the risk precursor index is less than or equal to the first preset threshold, it is determined that the monitoring node is in a normal evolution pattern, and no early warning is issued; In the embodiment, the first preset threshold can be set to 1.5; When the risk precursor index When the value is greater than the first preset threshold and less than or equal to the second preset threshold, the node's state is determined to have a statistically significant deviation, and a warning at the attention level is issued; in this embodiment, the second preset threshold can be set to 3.0; When the risk precursor index When the value exceeds the second preset threshold, the node's state is determined to have deviated significantly, and a risk warning is issued. The setting of the second preset threshold of 3.0 can be derived from the three sigma criterion in statistics. By establishing a threshold system with clear statistical significance, the automatic conversion from quantitative risk index to graded early warning is realized. Compared with the traditional method of relying on experience to set fixed physical quantity thresholds, this early warning logic has stronger adaptability and theoretical basis because the risk precursor index it evaluates is itself a normalized, dimensionless statistical quantity, which makes the early warning system more universal for landslides with different geological conditions.

[0033] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation, characterized in that, The method comprises the following steps: a data acquisition module is used to acquire multi-source heterogeneous sensor data streams of a landslide monitoring area, and based on the multi-source heterogeneous sensor data streams, a normalized feature vector representing the state of each monitoring node is constructed; an association modeling module is used to discretize the landslide monitoring area into heterogeneous nodes, and based on a weighted adjacency matrix containing geological prior knowledge, the physical influence degree between the heterogeneous nodes is quantified; a state prediction module is used to iteratively update the hidden state of each heterogeneous node based on the normalized feature vector and the weighted adjacency matrix, and the hidden state is interpreted as an incremental displacement; a risk assessment module is used to calculate the risk precursor index of each heterogeneous node deviating from the normal evolution mode based on the hidden state, and determine and issue a graded warning according to a preset threshold system.

2. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 1, characterized in that, The multi-source heterogeneous sensor data stream includes original interferometric phase stream data and coherence stream data, which are acquired by InSAR satellite; the multi-source heterogeneous sensor data stream also includes three-dimensional real-time displacement vector stream, which is acquired by GNSS station; the multi-source heterogeneous sensor data stream also includes pulse rainfall data, which is acquired by rain gauge.

3. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 1, characterized in that, The association modeling module is used to determine the weighted adjacency matrix by combining the Gaussian attenuation effect based on spatial distance and the topographic influence effect based on elevation difference; wherein the topographic influence effect simulates the asymmetric influence under the action of gravity through nonlinear transformation to enhance the influence of upstream nodes on downstream nodes.

4. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 1, characterized in that, The state prediction module iteratively updates the hidden state of each heterogeneous node, including: through graph convolution operation, the hidden state of each neighbor node at the last time is weighted aggregated to obtain the comprehensive influence representing the spatial neighborhood transmission; the comprehensive influence and the normalized feature vector of the heterogeneous node at the current time are jointly input into the gated recurrent unit to calculate the hidden state at the current time.

5. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 1, characterized in that, The spatio-temporal graph neural network model is trained by using a hybrid loss function; the hybrid loss function includes prediction accuracy loss and physical regularization term.

6. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 5, characterized in that, The physical regularization term is used to impose physical constraints during model training; wherein the physical constraint is that the deformation rate of the landslide should not slow down when effective rainfall occurs.

7. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 1, characterized in that, The risk assessment module calculates the risk precursor index, including: using stable historical monitoring data, a set of hidden state vectors representing normal state is generated by the trained spatio-temporal graph neural network model; based on the hidden state vectors representing normal state, the average hidden state and standard deviation of each node are calculated; by measuring the Euclidean distance between the hidden state at the current time and the average hidden state, and normalizing it with the standard deviation, the risk precursor index is obtained.

8. The complex high-order landslide risk dynamic identification system based on multi-source InSAR cooperation according to claim 1, characterized in that, The graded warning includes: when the risk precursor index is less than or equal to the first preset threshold, it is determined to be in normal state and no warning is issued; when the risk precursor index is greater than the first preset threshold and less than or equal to the second preset threshold, a concern level warning is issued; when the risk precursor index is greater than the second preset threshold, a risk level warning is issued.

Images