Accurate dynamic risk assessment method for dike piping danger

Abstract

The invention discloses an accurate dynamic risk assessment method for a dike piping dangerous case, and the method comprises the steps: obtaining multi-modal data, and extracting a feature vector of the multi-modal data; obtaining a multi-modal feature matrix according to the feature vector, and fusing the feature matrix to obtain a fused feature matrix; a dike piping risk assessment model is constructed, the fusion feature matrix is input into the dike piping risk assessment model, and the dike piping risk assessment model is obtained through training of a training set; judging whether the dike piping risk assessment result exceeds a preset threshold value or not, if yes, triggering an early warning mechanism, and if not, continuing to obtain the dike piping related multi-modal data, returning to the step 1 to repeatedly execute the subsequent steps, and continuously monitoring the change trend of the multi-modal data.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk assessment, and particularly relates to a precise dynamic risk assessment method for levee piping hazards. Background Technique

[0002] In recent years, researchers have proposed various new levee inspection methods based on a variety of new equipment platforms. The main new equipment platforms include helicopters, drones, robot dogs, manned vehicles, unmanned vehicles, etc. Xu Xianlei et al. integrated the ground penetrating radar originally used in geological environment surveys into a manned survey vehicle, vertically emitted high-frequency battery pulse waves (20 - 40 meters) downward from the top of the dam body, and determined the specific scope of factors inducing piping hazards such as bottom seepage water and pests in the dam body through image analysis. This method can detect hazards over a long distance along the levee, but the disadvantage is that the vehicle-mounted equipment is relatively large in volume and easily causes congestion on the levee road; Tang Liang et al. installed an infrared imager and a temperature sensing device on a bionic robot dog, used the thermal infrared instrument for video inspection, and used the temperature sensing device to measure the temperature of the piping point. This method can effectively be used to replace manual on-site discovery and verification of piping hazards, but it cannot be far from the command cabin, so the inspection range is small; Chen Xiaobin et al. integrated the magnetotelluric technology equipment onto a tripod, emitted radio frequency electromagnetic waves into the ground from 0 to 50 meters outside the levee slope to observe the impedance of underground substances, explored the channels where piping hazards occur, and determined the specific location through image analysis. This method can provide scientific and technological support for effectively eradicating piping hazards and has great development prospects, but currently, the operation time is long and the coverage range is small, and it cannot be promoted in actual combat. Based on the above problems, there is an urgent need for a precise dynamic risk assessment method for levee piping hazards. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a precise dynamic risk assessment method for levee piping hazards. By carrying detection devices such as visible light and thermal infrared on a drone platform, near-real-time data transmission and processing can be carried out to achieve rapid and large-scale inspection of levee hazards.

[0004] The present invention provides a precise dynamic risk assessment method for levee piping hazards, including:

[0005] Step 1, obtain multimodal data and extract the feature vectors of the multimodal data;

[0006] Step 2, according to the feature vectors, obtain a multimodal feature matrix, and fuse the feature matrix to obtain a fused feature matrix;

[0007] Step 3, construct a levee piping risk assessment model, input the fused feature matrix into the levee piping risk assessment model, and obtain the levee piping risk assessment result, wherein the levee piping risk assessment model is obtained by training with a training set;

[0008] Step 4. Determine whether the embankment piping risk assessment result exceeds a preset threshold. If it exceeds the preset threshold, trigger an early warning mechanism. If it does not exceed the preset threshold, continue to obtain multi-modal data related to embankment piping, return to Step 1 to repeat the subsequent steps, and continuously monitor the change trend of the multi-modal data.

[0009] Optionally, the multi-modal data includes: image data, text data, and sensor data.

[0010] Optionally, extracting the feature vectors of the multi-modal data includes:

[0011] Use a convolutional neural network to extract the spatial features of the image data and obtain an image feature vector;

[0012] Use natural language processing technology to extract the semantic features of the text data and obtain a text feature vector;

[0013] Use time series analysis methods to extract the temporal features of the sensor data and obtain a sensor feature vector.

[0014] Optionally, obtaining the multi-modal feature matrix according to the feature vectors includes:

[0015] Map the image feature vector, text feature vector, and sensor feature vector to a unified feature space respectively to obtain a unified represented feature vector;

[0016] For the mapped feature vectors, use a clustering algorithm to cluster the feature vectors, aggregate the feature vectors with high similarity together, and form multiple feature clusters;

[0017] For each feature cluster, determine the center point of the feature cluster by calculating the similarity between the feature vectors inside the feature cluster, and use it as the representative feature vector of the corresponding feature cluster;

[0018] Combine the representative feature vectors of each feature cluster to obtain the multi-modal feature matrix.

[0019] Optionally, fusing the feature matrix to obtain a fused feature matrix includes:

[0020] Use an attention mechanism to calculate the attention weights between different modal features in the multi-modal feature matrix to obtain an attention weight matrix;

[0021] According to the attention weight matrix, perform weighted fusion on the multi-modal feature matrix to obtain a fused feature matrix.

[0022] Optionally, obtaining the training set includes:

[0023] Obtain the historical data and real-time monitoring data of levee piping, preprocess the data, extract the characteristic parameters related to piping, and construct a piping characteristic matrix;

[0024] Construct a levee attribute characteristic matrix according to the levee structure parameters, soil parameters and hydrological parameters;

[0025] Fuse the levee attribute characteristic matrix and the piping characteristic matrix to obtain a training set.

[0026] Optionally, the levee piping risk assessment model includes: convolutional neural network, recurrent neural network, transformation network, graph neural network, autoencoder, generative adversarial network, reinforcement learning model, mixture of experts model or residual network.

[0027] Optionally, determining whether the levee piping risk assessment result exceeds a preset threshold includes:

[0028] Obtain evaluation indicators;

[0029] Use the AHP (Analytic Hierarchy Process) to obtain the weights of the evaluation indicators;

[0030] Perform weighted processing on the evaluation indicators according to the weights of the evaluation indicators, and perform dynamic risk assessment according to the weighted indicators to determine whether the levee piping risk assessment result exceeds a preset threshold.

[0031] Optionally, using the AHP to obtain the weights of the evaluation indicators includes:

[0032] Use the AHP to construct a hierarchical structure of the evaluation indicators;

[0033] Based on the hierarchical structure, construct a judgment matrix, and quantify the relative importance degree of the indicators according to the judgment matrix;

[0034] Calculate the subjective weights of the indicators, and test the consistency of the judgment matrix according to the subjective weights;

[0035] According to the test results, obtain the weights of the evaluation indicators.

[0036] Optionally, performing weighted processing on the evaluation indicators according to the weights of the evaluation indicators and performing dynamic risk assessment according to the weighted indicators includes:

[0037] Perform weighted processing on the evaluation indicators to obtain a hazard index, a sensitivity index and a vulnerability index;

[0038] According to the hazard index, the sensitivity index and the vulnerability index, obtain the risk index of levee piping;

[0039] Perform dynamic risk assessment based on the risk index of embankment piping.

[0040] Compared with the prior art, the present invention has the following advantages and technical effects:

[0041] With the present invention, a visible light and thermal infrared detection device is carried on a drone platform to perform near-real-time data transmission and processing, realizing rapid and large-scale inspection of embankment danger situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0043] Figure 1 is a flowchart of a method for precise dynamic risk assessment of embankment piping danger situations according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine with the embodiments to detail this application.

[0045] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0046] The drone platform has the characteristics of miniaturization, light weight, low cost, etc. A visible light and thermal infrared detection device carried on it can perform near-real-time data transmission and processing, realizing rapid and large-scale inspection of embankment danger situations.

[0047] If the drone is compared to a human body, then the visible light and thermal infrared payloads are equivalent to a person's eyes and limbs, and the intelligent recognition algorithm is equivalent to a person's brain. The "brain" judges the piping danger situation based on the image data collected by the "human body". Currently, the mainstream recognition algorithms mainly include two categories: image recognition technology based on machine learning and temperature anomaly recognition algorithms based on the principle of thermal imaging. Based on the above platform and technology, the present invention has gradually explored a set of efficient inspection methods for embankment danger situations with human-machine collaboration.

[0048] The present invention proposes a method for precise dynamic risk assessment of embankment piping danger situations, as Figure 1 shown, specifically including the following steps:

[0049] Step 1: Obtain multi-modal data and extract the feature vectors of the multi-modal data;

[0050] Step 2: Obtain a multi-modal feature matrix based on the eigenvectors, and fuse the feature matrix to obtain a fused feature matrix;

[0051] Step 3: Construct a dike piping risk assessment model, and input the fused feature matrix into the dike piping risk assessment model, which is obtained by training with a training set;

[0052] Step 4: Determine whether the dike piping risk assessment result exceeds a preset threshold. If it exceeds the preset threshold, trigger an early warning mechanism. If it does not exceed the preset threshold, continue to obtain multi-modal data related to dike piping, return to Step 1 and repeat the subsequent steps to continuously monitor the changing trend of the multi-modal data.

[0053] Specifically, construct a dike piping risk assessment model, input the fused feature matrix into the dike piping risk assessment model, and output the dike piping risk assessment result.

[0054] Obtain the historical data and real-time monitoring data of dike piping, preprocess the data, extract the feature parameters related to piping, and construct a piping feature matrix. Construct a dike attribute feature matrix based on the dike structure parameters, soil parameters, hydrological parameters, etc., and fuse it with the piping feature matrix. Input the fused feature matrix into the pre-constructed dike piping risk assessment model, and capture the dynamic change rules and trends of dike piping occurrence through network training. Use the trained dike piping risk assessment model to predict the risk of newly input dike piping monitoring data to obtain the piping risk assessment result. If the piping risk assessment result exceeds the preset threshold, warn the piping risk level of the dike and prompt to take corresponding prevention and control measures. Automatically generate a dike piping risk distribution map according to the dike piping risk assessment result to visually display the high-risk areas of piping. Continuously optimize and update the parameters of the dike piping risk assessment model to improve the accuracy and reliability of the dike piping risk assessment.

[0055] Input the fused feature matrix into the pre-constructed dike piping risk assessment model, and capture the dynamic change rules and trends of dike piping occurrence through network training.

[0056] According to the historical data of levee piping occurrence, multi-dimensional time-series data reflecting the characteristics of piping occurrence are extracted to construct a feature matrix. Data preprocessing techniques are used to clean and standardize the feature matrix, eliminating noise and outliers in the data and improving data quality. Through a feature selection algorithm, a feature subset highly correlated with the occurrence of piping is screened out, reducing the dimension of the feature matrix and redundant information. The optimized feature matrix is divided into a training set and a test set, where the training set is used for model training and the test set is used for model evaluation. A levee piping risk assessment model is constructed, and appropriate numbers of network layers and neurons are set according to the time-series characteristics of the feature matrix. The training set is input into the neural network model, and through forward propagation and backpropagation algorithms, the network weights are continuously adjusted to minimize the error between the model output and the actual piping occurrence. The trained neural network model is used to predict the test set, and according to the degree of coincidence between the prediction results and the actual situation, the generalization ability of the model is evaluated, and the model parameters are further optimized to improve the accuracy of piping prediction.

[0057] Using the trained levee piping risk assessment model, risk prediction is carried out on newly input levee piping monitoring data to obtain the piping risk assessment result.

[0058] Furthermore, the multi-modal data includes: image data, text data, and sensor data.

[0059] Specifically, for the image data in the multi-modal data, image preprocessing techniques are used to denoise, enhance, and standardize the images to obtain preprocessed image data. Among them, the image data includes: visible light data, thermal infrared data, and laser ranging, etc.; for the text data in the multi-modal data, natural language processing techniques are used to perform word segmentation, part-of-speech tagging, named entity recognition, etc. on the text to extract key information and obtain structured text data; for the sensor data in the multi-modal data, according to the sensor type and data characteristics, corresponding data preprocessing methods are used to denoise, interpolate, and normalize the sensor data to obtain standardized sensor data.

[0060] Furthermore, the extraction of feature vectors from multi-modal data includes:

[0061] A convolutional neural network is used to extract the spatial features of the image data to obtain image feature vectors;

[0062] Natural language processing techniques are used to extract the semantic features of the text data to obtain text feature vectors;

[0063] A time series analysis method is used to extract the time-series features of the sensor data to obtain sensor feature vectors.

[0064] Specifically, image data, text data, and sensor data are obtained, and corresponding feature extraction methods are adopted for different types of data. For image data, a convolutional neural network is used to extract its features, and the spatial features of the image are extracted through convolutional layers and pooling layers to obtain an image feature vector. For text data, natural language processing techniques are used to extract its features, and the semantic features of the text are extracted through word embedding and recurrent neural networks to obtain a text feature vector. For sensor data, a time series analysis method is used to extract its features, and the temporal features of the sensor data are extracted through a sliding window and statistical features to obtain a sensor feature vector.

[0065] Further, according to the feature vectors, a multi-modal feature matrix is obtained, including:

[0066] The image feature vector, text feature vector, and sensor feature vector are respectively mapped into a unified feature space to obtain a feature vector with unified representation;

[0067] For the mapped feature vectors, a clustering algorithm is used to cluster the feature vectors, and the feature vectors with high similarity are aggregated together to form multiple feature clusters;

[0068] For each feature cluster, by calculating the similarity between the feature vectors within the feature cluster, the center point of the feature cluster is determined as the representative feature vector of the corresponding feature cluster;

[0069] The representative feature vectors of each feature cluster are combined together to obtain a multi-modal feature matrix.

[0070] Specifically, for image features, a pre-trained convolutional neural network can be used to extract high-dimensional features, which are then mapped to a low-dimensional unified space through a fully connected layer. For text features, word vectors can be obtained through word embedding first, then encoded into sentence vectors using a recurrent neural network, and finally mapped to the unified space. Sensor data can be mapped after extracting temporal features using a temporal convolutional network. The choice of clustering algorithm is crucial for the formation of feature clusters. The K-means algorithm is suitable for spherical clusters, while the DBSCAN algorithm is more suitable for handling clusters with irregular shapes. Taking image recognition as an example, assuming there are feature vectors of 1000 pictures, the K-means algorithm can be used to cluster them into 10 clusters. Each cluster may represent a type of object, such as cats, dogs, cars, etc. Determining the center points of the feature clusters is to obtain the most representative features. The average value of all vectors within the cluster can be calculated as the center point, or the vector with the minimum average distance to other vectors can be selected as the center point. This can greatly reduce the amount of data and improve the efficiency of subsequent processing. Constructing a multi-modal feature matrix is to comprehensively utilize various modal information. For example, in a smart home scenario, the feature vectors of images (such as the pictures taken by cameras), text (such as the text transcribed from user voice commands), and sensor data (such as temperature and humidity) can be combined into a matrix. This can comprehensively describe the current home environment state. Principal component analysis (PCA) is used for dimensionality reduction, which can effectively remove redundant information and highlight key features. For example, a 100-dimensional feature vector may be reduced to 20 dimensions after PCA, but still retains 90% of the information. This not only reduces the computational amount but also suppresses noise and improves the generalization ability of the model.

[0071] Furthermore, fusing the feature matrix to obtain a fused feature matrix includes:

[0072] Adopting an attention mechanism to calculate the attention weights between different modal features in the multi-modal feature matrix to obtain an attention weight matrix;

[0073] According to the attention weight matrix, performing weighted fusion on the multi-modal feature matrix to obtain a fused feature matrix.

[0074] Furthermore, obtaining a training set includes:

[0075] Obtaining historical data and real-time monitoring data of dike piping, preprocessing the data, extracting piping-related feature parameters, and constructing a piping feature matrix;

[0076] According to the dike structure parameters, soil parameters, and hydrological parameters, constructing a dike attribute feature matrix;

[0077] Fusing the dike attribute feature matrix and the piping feature matrix to obtain a training set.

[0078] Furthermore, the levee piping risk assessment model includes: convolutional neural network, recurrent neural network, transformation network, graph neural network, autoencoder, generative adversarial network, reinforcement learning model, mixture of experts model, or residual network.

[0079] Furthermore, determining whether the levee piping risk assessment result exceeds a preset threshold includes:

[0080] Obtain evaluation indicators;

[0081] Using the AHP (Analytic Hierarchy Process), obtain the weights of the evaluation indicators;

[0082] Perform weighted processing on the evaluation indicators according to the weights of the evaluation indicators, and conduct dynamic risk assessment based on the weighted indicators to determine whether the levee piping risk assessment result exceeds the preset threshold.

[0083] Specifically, the weights of each level of evaluation indicators are generally obtained by the Analytic Hierarchy Process (AHP). The Analytic Hierarchy Process is a process of modeling and quantifying the decision-making thinking process of decision-makers for complex systems. By decomposing complex problems into several levels and several factors, and making simple comparisons and calculations among the factors, the weights of the importance degrees of different solutions can be obtained, providing a basis for the selection of the best solution. In a broad sense, disaster risk assessment is to conduct risk assessment on a disaster system, that is, on the basis of separately conducting risk assessment on the disaster-forming environment, hazard-causing factors, and disaster-bearing bodies, conduct risk assessment on the disaster system. In the present invention, the levee piping disaster risk is a comprehensive function of three aspects: the danger of hazard-causing factors, the sensitivity of the disaster-forming environment, and the vulnerability of disaster-bearing bodies. The levee piping disaster risk index can be calculated by a formula.

[0084] Furthermore, using the AHP to obtain the weights of the evaluation indicators includes:

[0085] Use the AHP to construct a hierarchical structure of the evaluation indicators;

[0086] Based on the hierarchical structure, construct a judgment matrix, and quantify the relative importance degree of the indicators according to the judgment matrix;

[0087] Calculate the subjective weights of the indicators, and test the consistency of the judgment matrix according to the subjective weights;

[0088] According to the test results, obtain the weights of the evaluation indicators.

[0089] Specifically, first, establish a hierarchical structure. Second, construct pairwise judgment matrices, and use the scale method of 1-9 and their reciprocals to quantify the relative importance of pairwise elements. Third, calculate the subjective weight Wi of the indicators. Fourth, conduct a consistency test. When CR < 0.1, it indicates that the consistency of the judgment matrix is reasonable; when CR ≥ 0.1, it means that the judgment matrix is unreasonable and needs to be re-tested for consistency.

[0090] Furthermore, perform a weighted process on the evaluation indicators according to their weights, and conduct a risk assessment based on the weighted indicators, including:

[0091] Perform a weighted process on the evaluation indicators to obtain the hazard index, sensitivity index, and vulnerability index;

[0092] Obtain the risk index of levee piping according to the hazard index, sensitivity index, and vulnerability index;

[0093] Conduct a risk assessment based on the risk index of levee piping.

[0094] Furthermore, the method for performing a weighted process on the evaluation indicators to obtain the hazard index, sensitivity index, and vulnerability index is as follows:

[0095]

[0096] Among them, H(x) is the hazard index, S(x) is the sensitivity index, V(x) is the vulnerability index, H ji (x), S ji (x), V ji (x) are the values after standardizing each indicator, and w j is the weight of each indicator;

[0097] The method for obtaining the risk index of levee piping is as follows:

[0098] R(x) = w H H(x) × w S S(x) × w V V(x)

[0099] Among them, R(x) is the risk index, and w H , w S , w V are the weights of the hazard index, sensitivity index, and vulnerability index, respectively.

[0100] Specifically, the proportion relationship between each indicator and criterion is shown in Table 1:

[0101]

[0102] Among them, the indicator weights are dynamically changing.

[0103] Specifically, it is determined whether the embankment piping risk assessment result exceeds a preset threshold. If it exceeds the preset threshold, the early warning mechanism is triggered. If it does not exceed the preset threshold, multi-modal data related to embankment piping is continuously acquired, and the process returns to step 1 to repeat the subsequent steps, continuously monitoring the change trend of the multi-modal data.

[0104] Continuously acquire multi-modal data related to embankment piping through a sensor network, including parameters such as seepage flow rate, seepage pressure, and soil displacement, to form a real-time data stream. Preprocess the acquired multi-modal data to remove outliers and noise data, obtaining a normalized data set. According to a pre-established embankment piping risk assessment model, use machine learning algorithms such as support vector machines or random forests to perform a piping risk assessment on the normalized data set, obtaining a risk assessment result. Determine whether the risk assessment result exceeds a preset threshold. If it exceeds the threshold, trigger the early warning mechanism, send a warning message to the management personnel, and at the same time automatically adjust the monitoring frequency and data acquisition strategy. If the risk assessment result does not exceed the preset threshold, continue to acquire multi-modal data related to embankment piping, update the real-time data stream, and return to step 2 to repeat the data preprocessing and risk assessment process. During the continuous monitoring process, use an incremental learning method to continuously optimize and update the piping risk assessment model, improving the accuracy and real-time performance of risk prediction. Implement an intuitive display of the embankment piping risk through visualization technology, generating a heat map or risk distribution map according to the risk level to assist the management personnel in making decisions and emergency responses.

[0105] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A precise dynamic risk assessment method for dike piping hazards, characterized in that: include: Step 1: Acquire multimodal data and extract feature vectors of the multimodal data; Step 2: obtaining a multimodal feature matrix according to the feature vector, and fusing the feature matrix to obtain a fused feature matrix; Step 3: construct a dike piping risk assessment model, input the fusion feature matrix into the dike piping risk assessment model, and obtain the dike piping risk assessment result, wherein the dike piping risk assessment model is obtained by training with a training set; Step 4: Determine whether the levee piping risk assessment result exceeds a preset threshold. If it exceeds the preset threshold, trigger the early warning mechanism. If it does not exceed the preset threshold, continue to obtain multimodal data related to the levee piping, return to step 1, repeat the subsequent steps, and continuously monitor the changing trend of the multimodal data.

2. According to claim 1, a method for accurate dynamic risk assessment of dike piping danger is characterized in that: The multimodal data includes: image data, text data and sensor data.

3. According to claim 2, a method for accurate dynamic risk assessment of dike piping danger is characterized in that: Extracting the feature vector of the multimodal data includes: Use convolutional neural network to extract spatial features of image data and obtain image feature vectors; Use natural language processing technology to extract the semantic features of text data and obtain text feature vectors; The time series analysis method is used to extract the time series characteristics of sensor data and obtain the sensor feature vector.

4. According to claim 3, a method for accurate dynamic risk assessment of dike piping danger is characterized in that: According to the feature vector, obtaining a multimodal feature matrix includes: Mapping the image feature vector, the text feature vector and the sensor feature vector to a unified feature space respectively to obtain a feature vector of a unified representation; For the mapped feature vectors, a clustering algorithm is used to cluster the feature vectors, and feature vectors with high similarity are aggregated together to form multiple feature clusters; For each feature cluster, the center point of the feature cluster is determined by calculating the similarity between the feature vectors within the feature cluster, which is used as the representative feature vector of the corresponding feature cluster; The representative feature vectors of each feature cluster are combined together to obtain the multimodal feature matrix.

5. According to claim 4, a method for accurate dynamic risk assessment of dike piping danger is characterized in that: Fusing the feature matrix to obtain a fused feature matrix includes: Using an attention mechanism, calculating the attention weights between different modal features in the multimodal feature matrix to obtain an attention weight matrix; According to the attention weight matrix, the multimodal feature matrix is ​​weightedly fused to obtain a fused feature matrix.

6. According to claim 1, a method for accurate dynamic risk assessment of dike piping danger is characterized in that: Acquiring the training set includes: Obtain historical data and real-time monitoring data of levee piping, pre-process the data, extract piping-related characteristic parameters, and construct a piping characteristic matrix; According to the levee structure parameters, soil parameters and hydrological parameters, the levee attribute characteristic matrix is ​​constructed; The levee attribute feature matrix and the piping feature matrix are fused to obtain the training set.

7. The accurate dynamic risk assessment method for dike piping danger according to claim 6 is characterized in that: The dike piping risk assessment model includes: a convolutional neural network, a recurrent neural network, a transformation network, a graph neural network, an autoencoder, a generative adversarial network, a reinforcement learning model, a hybrid expert model or a residual network.

8. The accurate dynamic risk assessment method for dike piping danger according to claim 7 is characterized in that: Determining whether the dike piping risk assessment result exceeds a preset threshold includes: Obtain evaluation indicators; Using AHP hierarchical analysis method, the weight of the evaluation index is obtained; The evaluation indicators are weighted according to their weights, and a dynamic risk assessment is performed based on the weighted indicators to determine whether the levee piping risk assessment result exceeds a preset threshold.

9. The accurate dynamic risk assessment method for dike piping danger according to claim 8 is characterized in that: Using the AHP hierarchical analysis method, the weights of the evaluation indicators are obtained, including: Using AHP analytic hierarchy process to construct the hierarchical structure of the evaluation index; Based on the hierarchical structure, a judgment matrix is ​​constructed, and the relative importance of the indicators is quantified according to the judgment matrix; Calculating the subjective weight of the indicator, and testing the consistency of the judgment matrix according to the subjective weight; According to the test result, the weight of the evaluation index is obtained.

10. The accurate dynamic risk assessment method for dike piping danger according to claim 8 is characterized in that: The evaluation indicators are weighted according to their weights, and dynamic risk assessment is performed according to the weighted indicators, including: Performing weighted processing on the evaluation indicators to obtain a hazard index, a sensitivity index and a vulnerability index; Obtaining a risk index of embankment piping according to the hazard index, sensitivity index and vulnerability index; A dynamic risk assessment is performed based on the risk index of levee piping.

Images