Dam safety monitoring method and device, computer equipment and storage medium

Abstract

The invention provides a dam safety monitoring method and device, computer equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing of collected multi-source monitoring time series data, inputting the data into a trained anomaly detection model, obtaining the anomaly probability distribution of each monitoring node, and obtaining the anomaly probability distribution of each monitoring node; and abnormal information is displayed through a dam safety monitoring visualization system. According to the method, the undirected graph is constructed through the propagation model of each monitoring type and the spatial features and the time features of the monitoring nodes, the undirected graph is input into the graph neural network for coding to obtain the second feature vector representing the state information of the dam structure, and the environment information is input into the first deep neural network for coding to obtain the first feature vector; and the first feature vector and the second feature vector are fused and then input to the second deep neural network to obtain the anomaly probability distribution, so that the accuracy and precision of dam anomaly detection are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of dam safety monitoring, and in particular relates to a dam safety monitoring method, device, computer equipment and storage medium. Background Art

[0002] In recent years, the automation of dam safety monitoring equipment has enabled people to obtain a large amount of dam monitoring data, and thus more and more data-driven methods are used for dam safety monitoring. Dam safety monitoring models mainly include statistical models and machine learning models. There are many types of statistical models, the most common of which are Kalman filter statistical models, logistic regression models, etc. Machine learning models include support vector machine SVM, random forest, Adaboost, Boostrap and other integrated learning models. The accuracy of statistical models is not as good as that of machine learning models, but the interpretability of machine learning models is not as good as that of statistical models.

[0003] With the rapid development of artificial intelligence, more and more neural network models and deep learning models are applied to dam safety monitoring, such as BP neural network and long short-term memory neural network model. These models mainly consider the inherent characteristics of the dam such as deformation, stress, strain, etc. for dam safety monitoring. Due to the small number of characteristic factors considered, the accuracy of dam anomaly detection is not high and the precision is poor. Therefore, a dam safety monitoring method is urgently needed to improve the accuracy and precision of dam safety monitoring. Summary of the invention

[0004] In view of the defects of the prior art and in order to improve the accuracy and visualization of dam safety monitoring, the present invention provides a dam safety monitoring method, device, computer equipment and storage medium, which can effectively solve the above problems.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] The present invention provides a dam safety monitoring method, comprising the following steps:

[0007] S1: Collect multi-source monitoring time series data, wherein the multi-source monitoring time series data includes dam structure status information and environmental information; wherein the dam structure status information includes stress, strain, deformation, seepage pressure and flow; and the environmental information includes one or more of reservoir water level, temperature, air pressure, rainfall and evaporation;

[0008] S2: preprocessing the multi-source monitoring time series data to obtain the multi-source monitoring time series data to be processed; the multi-source monitoring time series data to be processed includes the dam structure state information to be processed and the environmental information to be processed;

[0009] S3: Input the multi-source monitoring time series data to be processed into the trained anomaly detection model to identify and obtain dam anomaly information, including:

[0010] The anomaly detection model includes a first deep neural network, a graph neural network, and a second deep neural network;

[0011] Encoding the to-be-processed multi-source monitoring time series data using the first deep neural network and the graph neural network to obtain a third feature vector;

[0012] Decoding the third feature vector using the second deep neural network to obtain dam abnormality information;

[0013] S4: Based on the dam safety monitoring visualization system, the abnormal information of the dam is displayed.

[0014] Furthermore, the using the first deep neural network and the graph neural network to encode the multi-source monitoring time series data to be processed to obtain a third feature vector specifically includes:

[0015] An undirected graph is constructed using the propagation model of the dam structure status information and the location information of the monitoring nodes;

[0016] Inputting the environmental information to be processed into a first deep neural network to obtain a first feature vector;

[0017] Input the undirected graph into the graph neural network to obtain the second eigenvector;

[0018] Merging the first eigenvector and the second eigenvector to obtain a third eigenvector;

[0019] The using the second deep neural network to decode the third feature vector to obtain dam abnormality information specifically includes:

[0020] The third feature vector is input into the second deep neural network to obtain the abnormal probability distribution of the monitoring node.

[0021] Furthermore, the use of the propagation model of the dam structure status information and the monitoring node location information to construct an undirected graph specifically includes:

[0022] Using a stress propagation model, a strain propagation model, a deformation propagation model, a seepage pressure propagation model, and a flow propagation model, the properties of all monitoring nodes are obtained, wherein the properties of the monitoring nodes include stress, strain, deformation, seepage pressure, and flow;

[0023] Obtain the attributes of all monitoring nodes at time t, connect adjacent nodes by edge according to the spatial position of the monitoring nodes, and construct an undirected spatial subgraph; construct a spatial weight matrix W according to the distance between the monitoring nodes in the undirected spatial subgraph S , construct the spatial node attribute matrix X according to the stress, strain, deformation, seepage pressure and flow of the monitoring node S ;

[0024] Obtain the attributes of all monitoring nodes at time t-t1, t, and t+t2, connect adjacent nodes by edges according to the time distribution, and construct an undirected time subgraph; construct a time weight matrix W according to the time step in the undirected time subgraph T , construct the time node attribute matrix X according to the stress, strain, deformation, seepage pressure and flow of the monitoring node T .

[0025] Furthermore, the step of inputting the undirected graph into the graph neural network to obtain the second eigenvector specifically includes:

[0026] The graph neural network includes a first encoder, a second encoder and an aggregation layer;

[0027] Construct a spatial adjacency matrix A from an undirected spatial subgraph S ;

[0028] Construct the temporal adjacency matrix A from the undirected temporal subgraph T ;

[0029] The spatial adjacency matrix A S , spatial node attribute matrix X S , spatial weight matrix W S Input to the first encoder to obtain the spatial hidden vector H S :

[0030]

[0031] Among them, σ represents the activation function, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0032] The time adjacency matrix A T , time node attribute matrix X T , time weight matrix W T Input to the second encoder to obtain the time hidden vector H T ;

[0033]

[0034] in, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0035] H S and the time hidden vector H T Input to the aggregation layer to obtain the second eigenvector H (2) :H (2) =aggregate(H S ,H T ), where aggregate(.) represents the aggregation function.

[0036] Further, the step of inputting the environmental information to be processed into a first deep neural network to obtain a first feature vector; and fusing the first feature vector and the second feature vector to obtain a third feature vector specifically includes:

[0037] The first deep neural network adopts a bidirectional long short-term memory neural network; the environmental information to be processed is input into the bidirectional long short-term memory neural network for encoding to obtain a first feature vector H (1) ;

[0038] Use the following formula to convert the first eigenvector H (1) and the second eigenvector H (2) Splice and get the third eigenvector H (3) :H (3) =concat(H (1) ,H (2) ).

[0039] Further, the third feature vector is input into the second deep neural network to obtain the abnormal probability distribution of the monitoring node, specifically including:

[0040] The second deep neural network adopts a recurrent neural network model;

[0041] The third eigenvector H (3) Input to the recurrent neural network model and output the abnormal probability distribution P of the monitoring node: P = RNN (H (3) ).

[0042] Furthermore, it also includes: constructing a dam safety monitoring visualization system, wherein the dam safety monitoring visualization system includes an operation tree, a central window, and an auxiliary window;

[0043] The dam safety monitoring visualization system is used to display dam abnormality information, specifically including:

[0044] The monitoring nodes whose probability exceeds the preset threshold are regarded as abnormal monitoring nodes;

[0045] In the digital dam structure model, abnormal monitoring nodes are displayed in a second display mode, and normal monitoring nodes are displayed in a first display mode; the second display mode is different from the first display mode;

[0046] In the digital dam structure model, click on a monitoring node to view the monitoring data table and / or results diagram associated with the monitoring node in a pop-up window.

[0047] The present invention also provides a dam safety monitoring device, comprising the following modules:

[0048] A collection module is used to collect multi-source monitoring time series data, wherein the multi-source monitoring time series data includes dam structure status information and environmental information; wherein the dam structure status information includes stress, strain, deformation, seepage pressure and flow; and the environmental information includes one or more of reservoir water level, air temperature, air pressure, rainfall and evaporation;

[0049] A preprocessing module, used for preprocessing the multi-source monitoring time series data to obtain the multi-source monitoring time series data to be processed; the multi-source monitoring time series data to be processed includes the dam structure state information to be processed and the environmental information to be processed;

[0050] An anomaly detection module is used to input the multi-source monitoring time series data to be processed into a trained anomaly detection model to identify and obtain dam anomaly information; the anomaly detection model includes a first deep neural network, a graph neural network, and a second deep neural network; the multi-source monitoring time series data to be processed is encoded using the first deep neural network and the graph neural network to obtain a third feature vector, and the third feature vector is decoded using the second deep neural network to obtain the dam anomaly information;

[0051] The visualization module is used to display the abnormal information of the dam based on the dam safety monitoring visualization system.

[0052] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0053] The present invention also provides a computer-readable storage medium for storing computer instructions, wherein the computer instructions implement the steps of the above-mentioned method when executed by a processor.

[0054] The dam safety monitoring method, device, computer equipment and storage medium provided by the present invention have the following advantages:

[0055] (1) When performing anomaly detection on dam monitoring data, the present invention constructs an anomaly detection model including a first deep neural network, a graph neural network, and a second deep neural network. The environmental characteristics of the environment in which the dam is located are extracted by the first deep neural network, and the inherent characteristics of the dam structure state information are extracted by the graph neural network. The environmental characteristics and the inherent characteristics are fused and input into the second deep neural network, and the dam anomaly information is decoded to obtain the dam anomaly information. Since the environmental characteristics and the inherent characteristics are comprehensively considered, the accuracy of dam anomaly detection is improved.

[0056] (2) When obtaining the inherent characteristics of the dam structure status information, the present invention first uses the propagation model to obtain the stress, strain, deformation, seepage pressure and flow value of all monitoring nodes, and uses the value as the attribute of the node in the undirected graph, providing a solid data foundation for dam anomaly detection; at the same time, the present invention constructs an undirected spatial subgraph based on the relative positions of all monitoring nodes at the same time, and constructs a spatial weight matrix according to the distance between the monitoring nodes; constructs an undirected time subgraph based on the monitoring nodes at different times, constructs a time weight matrix according to the time step, and uses the stress, strain, deformation, seepage pressure and flow value of the monitoring node as the attribute of the monitoring node, and then encodes the undirected subgraph through the graph neural network to obtain the inherent characteristics of the dam status information. Since the distance characteristics between the monitoring nodes, the time series characteristics of the monitoring data, and the propagation characteristics of the monitoring data are taken into account, the inherent characteristics obtained by encoding are more in line with the monitoring node, further improving the accuracy of dam anomaly detection.

[0057] (3) When the present invention uses a graph neural network to encode an undirected graph, the hidden vector of the spatial dimension is obtained by the first encoder, and the hidden vector of the time dimension is obtained by the second encoder. Then, the hidden vector of the spatial dimension and the hidden vector of the time dimension are aggregated to obtain the feature vector of the monitoring node. Since the hidden vectors of the time dimension and the spatial dimension are aggregated, the inherent characteristics obtained by the encoding have spatial characteristics and temporal characteristics, which once again improves the accuracy of dam anomaly detection.

[0058] (4) The present invention constructs an integrated visualization system for associating and displaying abnormal monitoring points, abnormal information and monitoring data in a digital dam spatial structure model, thereby improving the timeliness and operation and maintenance capabilities of dam safety monitoring and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of a dam safety monitoring method provided by an embodiment of the present invention.

[0060] Figure 2 Schematic diagram of an anomaly detection model provided by an embodiment of the present invention.

[0061] Figure 3It is a schematic diagram of a dam safety monitoring visualization system provided by an embodiment of the present invention.

[0062] Figure 4 It is a schematic diagram of a dam safety monitoring device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] Reference Figure 1 The embodiment of the present invention provides a dam safety monitoring method, comprising the following steps:

[0065] S1: Collect multi-source monitoring time series data, wherein the multi-source monitoring time series data includes dam structure status information and environmental information; wherein the dam structure status information includes at least stress, strain, deformation, seepage pressure and flow; and the environmental information includes one or more of reservoir water level, air temperature, air pressure, rainfall and evaporation;

[0066] Different types of monitoring instruments are deployed on the monitoring section of the dam, such as displacement meters, strain gauges, stress gauges, seepage pressure gauges, flow meters, etc., to collect structural status information of the dam, such as deformation, stress, strain, seepage pressure, flow, etc.; resistance thermometers, optical fiber sensors, miniature multifunctional meteorological stations, etc. are deployed to collect dam environmental information such as reservoir water level, air temperature, air pressure, rainfall, evaporation, etc., and the collected dam structural status information and environmental information are uploaded to the server through the corresponding interface, so that the server can process the collected data.

[0067] During the collection process, the user can set the collection start time and collection end time through the interface, and upload the collected time series data to the server in a fixed period in the form of data packets.

[0068] S2: preprocessing the multi-source monitoring time series data to obtain the multi-source monitoring time series data to be processed; the multi-source monitoring time series data to be processed includes the dam structure state information to be processed and the environmental information to be processed;

[0069] The preprocessing of multi-source monitoring time series data usually includes missing value filling and data standardization. Missing value filling mainly uses missing value filling methods to fill in the missing data in the monitoring time series data. Missing value filling methods include mean method, median method or linear regression method.

[0070] Data standardization uses the min-max standardization method or the z-score standardization method. Assume that the collected stress time series data is C = {c1, c2, c3, ..., c T},c i ∈R N , using the min-max standardization method, the formula is:

[0071]

[0072] Among them, c' i Indicates c i Standardized data;

[0073] The z-score standardization method is used, and the calculation formula is:

[0074]

[0075] in, represents the mean of C, and σ represents the variance.

[0076] S3: Input the multi-source monitoring time series data to be processed into the trained anomaly detection model to identify and obtain dam anomaly information.

[0077] The anomaly detection model includes a first deep neural network, a graph neural network, and a second deep neural network; the multi-source monitoring time series data to be processed is encoded using the first deep neural network and the graph neural network to obtain a third feature vector; the third feature vector is decoded using the second deep neural network to obtain dam anomaly information;

[0078] The specific implementation method of this step includes:

[0079] S3.1: Construct an undirected graph using the propagation model of the dam structure status information and the location information of the monitoring nodes;

[0080] Using a stress propagation model, a strain propagation model, a deformation propagation model, a seepage pressure propagation model, and a flow propagation model, the properties of all monitoring nodes are obtained, wherein the properties of the monitoring nodes include stress, strain, deformation, seepage pressure, and flow;

[0081] Obtain the attributes of all monitoring nodes at time t, connect adjacent nodes by edge according to the spatial position of the monitoring nodes, and construct an undirected spatial subgraph; construct a spatial weight matrix W according to the distance between the monitoring nodes in the undirected spatial subgraph S , construct the spatial node attribute matrix X according to the stress, strain, deformation, seepage pressure and flow of the monitoring node S ;

[0082] Obtain the attributes of all monitoring nodes at time t-t1, t, and t+t2, connect adjacent nodes by edges according to the time distribution, and construct an undirected time subgraph; construct a time weight matrix W according to the time step in the undirected time subgraph T , construct the time node attribute matrix X according to the stress, strain, deformation, seepage pressure and flow of the monitoring node T .

[0083] S3.2: Inputting the environmental information to be processed into the first deep neural network to obtain a first feature vector;

[0084] The first deep neural network adopts a bidirectional long short-term memory neural network; the environmental information to be processed is input into the bidirectional long short-term memory neural network for encoding to obtain a first feature vector H (1) ;

[0085] S3.3: Input the undirected graph into the graph neural network to obtain the second eigenvector;

[0086] The graph neural network includes a first encoder, a second encoder and an aggregation layer;

[0087] Construct a spatial adjacency matrix A from an undirected spatial subgraph S ;

[0088] Construct the temporal adjacency matrix A from the undirected temporal subgraph T ;

[0089] The spatial adjacency matrix A S , spatial node attribute matrix X S , spatial weight matrix W S Input to the first encoder to obtain the spatial hidden vector H S :

[0090]

[0091] Among them, σ represents the activation function, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0092] The time adjacency matrix A T , time node attribute matrix X T , time weight matrix W T Input to the second encoder to obtain the time hidden vector H T ;

[0093]

[0094] in, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0095] H S and the time hidden vector H T Input to the aggregation layer to obtain the second eigenvector H (2) :H (2) =aggregate(H S ,H T ), where aggregate(.) represents the aggregation function.

[0096] S3.4: Fusing the first feature vector and the second feature vector to obtain a third feature vector;

[0097] Use the following formula to convert the first eigenvector H (1) and the second eigenvector H (2) Splice and get the third eigenvector H (3) :H (3) =concat(H (1) ,H (2) ).

[0098] S3.5: Input the third feature vector into the second deep neural network to obtain the abnormal probability distribution of the monitoring node.

[0099] The second deep neural network adopts a recurrent neural network model; the third feature vector H (3) Input to the recurrent neural network model and output the abnormal probability distribution P of the monitoring node: P = RNN (H (3) ).

[0100] The following is a detailed description of step S3:

[0101] like Figure 2 The anomaly detection model proposed in the present invention includes a first deep neural network, a graph neural network and a second deep neural network. The first deep neural network is used to encode environmental information such as reservoir water level, temperature, air pressure, rainfall and / or evaporation in the monitoring data to obtain a first eigenvector; the graph neural network is used to encode an undirected graph constructed according to the monitoring points and dam structure state information such as stress, strain, deformation, seepage pressure and flow in the monitoring data to obtain a second eigenvector. The second deep neural network is used to decode a third eigenvector obtained by concatenating the first eigenvector and the second eigenvector to obtain a probability distribution of whether the monitoring point is abnormal.

[0102] Since different monitoring nodes are equipped with different types of monitoring instruments, the original processed data collected by different monitoring nodes do not cover all types of dam structure status information. Therefore, it is necessary to use the processed data and propagation model to obtain all types of dam structure status information of the monitoring nodes.

[0103] Before using the processed data and propagation model to obtain all types of dam structure status information of monitoring nodes, all monitoring nodes are clustered, and stress monitoring nodes are used as the first monitoring node set, strain monitoring nodes are used as the second monitoring node set, deformation monitoring nodes are used as the third monitoring node set, seepage pressure monitoring nodes are used as the fourth monitoring node set, and flow monitoring nodes are used as the fifth monitoring node set.

[0104] For example, a total of 12 monitoring nodes AL are arranged on the dam, monitoring nodes A, G, and L are used to collect stress data, monitoring nodes B and E are used to collect deformation data, monitoring nodes C, F, and K are used to collect strain data, monitoring nodes H and J are used to collect flow data, and monitoring nodes D and I are used to collect seepage pressure data. Then A, G, and L are used as the first monitoring node set, C, F, and K are used as the second monitoring node set, B and E are used as the third monitoring node set, D and I are used as the fourth monitoring node set, and H and J are used as the fifth monitoring node set.

[0105] The stress propagation model is used to obtain the stress of the second monitoring node set, the third monitoring node set, the fourth monitoring node set and the fifth monitoring node set. The measured stress data in the first monitoring node set is input into the stress propagation model, and the stress of each monitoring node in the second monitoring node set, the third monitoring node set, the fourth monitoring node set and the fifth monitoring node set is predicted by the stress propagation model. For example, the data to be processed collected at the monitoring nodes A, G and L are input into the stress propagation model to obtain the stress at C, F, K, B, E, D, I, H and J. Among them, the stress propagation model can adopt a logistic regression model, a linear regression model, a deep neural network model trained based on the historically collected stress data and the monitoring node position, or a classic finite element model can be used to predict the stress at other monitoring nodes using the finite element model and the measured stress data.

[0106] The strain propagation model is used to obtain the strains of the nodes in the first monitoring node set, the third monitoring node set, the fourth monitoring node set, and the fifth monitoring node set; the strain data measured in the second monitoring node set is input into the strain propagation model, and the strain of each monitoring node in the first monitoring node set, the third monitoring node set, the fourth monitoring node set, and the fifth monitoring node set is predicted by the strain propagation model. For example, the data to be processed collected at the monitoring nodes C, F, and K are input into the strain propagation model to obtain the strains at A, G, L, B, E, D, I, H, and J. The strain propagation model can adopt a logistic regression model, a linear regression model, or a deep neural network model trained based on historically collected strain data and monitoring node positions, or a classic finite element model can be used to predict the stress at other monitoring nodes using the finite element model and the measured strain data.

[0107] The deformation propagation model is used to obtain the deformation of the nodes at the first monitoring node set, the second monitoring node set, the fourth monitoring node set and the fifth monitoring node set. The measured deformation data in the third monitoring node set is input into the deformation propagation model, and the deformation of each monitoring node in the first monitoring node set, the second monitoring node set, the fourth monitoring node set and the fifth monitoring node set is predicted by the deformation propagation model. For example, the unprocessed data collected at monitoring nodes B and E are input into the deformation propagation model to obtain the deformation at A, G, L, C, F, K, D, I, H and J. Among them, the deformation propagation model can adopt a logistic regression model, a linear regression model, a deep neural network model trained based on historically collected deformation data and monitoring node positions, or a classic finite element model can be used to predict the deformation at other monitoring nodes using the finite element model and the measured deformation data.

[0108] The osmotic pressure propagation model is used to obtain the osmotic pressure of the nodes at the first monitoring node set, the second monitoring node set, the third monitoring node set and the fifth monitoring node set. The measured osmotic pressure data in the fourth monitoring node set is input into the osmotic pressure propagation model, and the osmotic pressure of each monitoring node in the first monitoring node set, the second monitoring node set, the third monitoring node set and the fifth monitoring node set is predicted by the osmotic pressure propagation model. For example, the unprocessed data collected at the monitoring nodes D and I are input into the osmotic pressure propagation model to obtain the osmotic pressure at A, G, L, C, F, K, B, E, H and J. Among them, the osmotic pressure propagation model can adopt a logistic regression model, a linear regression model, a deep neural network model based on the historically collected osmotic pressure data and the monitoring node position training, or a classic finite element model can be used to predict the osmotic pressure at other monitoring nodes using the finite element model and the measured osmotic pressure data.

[0109] The traffic propagation model is used to obtain the traffic at the nodes of the first monitoring node set, the second monitoring node set, the third monitoring node set and the fourth monitoring node set. The measured traffic data in the fifth monitoring node set is input into the traffic propagation model, and the traffic of each monitoring node in the first monitoring node set, the second monitoring node set, the third monitoring node set and the fourth monitoring node set is predicted by the traffic propagation model. For example, the data to be processed collected at monitoring nodes H and J are input into the traffic propagation model to obtain the traffic at A, G, L, C, F, K, B, E, D, and I. Among them, the traffic propagation model can adopt a logistic regression model, a linear regression model, a deep neural network model trained based on historically collected traffic data and monitoring node positions, or a classic finite element model can be used to predict the traffic at other monitoring nodes using the finite element model and the measured traffic data.

[0110] The undirected graph constructed using Indicates, where V represents the monitoring node, E represents the edge, and X represents the monitoring node attribute. That is, X = {(x1, x2, x3, x4, x5)}, x1, x2, x3, x4, x5 respectively represent the values ​​of stress, strain, deformation, seepage pressure and flow at the monitoring node after standardization or calculation by the propagation model. Since the dam structure status information has a time series characteristic, when constructing an undirected graph, an undirected spatial subgraph is first constructed based on the spatial position. Constructing undirected temporal subgraph based on temporal distribution

[0111] In constructing an undirected spatial subgraph When t is reached, the monitoring node attributes of all monitoring nodes at time t are obtained, and the adjacent nodes are connected by edges according to the spatial position to construct an undirected spatial subgraph. The weight matrix W is constructed according to the distance between the monitoring nodes. S The stress, strain, deformation, seepage pressure and flow value corresponding to the monitoring node obtained by standardization or propagation model calculation are used as the spatial node attribute matrix X S For example, if monitoring node A is adjacent to monitoring nodes B and F in space, then edge connections are established between monitoring nodes A and B, and between monitoring nodes A and F, and the distance between A and B is used as the weight of the AB edge, and the distance between A and F is used as the weight of the AF edge. The spatial weight matrix is ​​constructed based on the weights of the AB edge and the AF edge. The standardized stress value collected by monitoring node A And the strain calculated by the propagation model Deformation Osmotic pressure flow As the spatial node attribute matrix of monitoring node A

[0112] In constructing an undirected temporal subgraph When t-t1, t, and t+t2 are used, the monitoring node attributes of all monitoring nodes at time t-t1, t, and t+t2 are obtained, and the adjacent nodes are connected by edges according to the time distribution to construct an undirected time subgraph. The time weight matrix W is constructed according to the time step T The stress, strain, deformation, seepage pressure and flow value corresponding to the monitoring node obtained by standardization or propagation model calculation are used as the time node attribute matrix X T For example, according to the time distribution, all monitoring nodes at time t-t1, t, t+t2 are obtained, and the monitoring node A at time t-t1, t, t+t2 is connected by edges, and the weight of the edge is constructed according to the time step t1 and t2. The standardized stress value collected by monitoring node A And the strain calculated by the propagation model Deformation Osmotic pressure flow As the time node attribute matrix of monitoring node A Connect the monitoring node B at time t-t1, t, and t+t2, and construct the edge weight according to the time steps t1 and t2 The standardized strain value collected by monitoring node B And the strain calculated by the propagation model Deformation Osmotic pressure flow As the time node attribute matrix of monitoring node B And so on.

[0113] After the undirected graph is constructed, it is input into the graph neural network to obtain the second eigenvector. Since the constructed undirected graph contains an undirected time subgraph and an undirected space subgraph, two encoders are required to encode the undirected time subgraph and the undirected space subgraph respectively to obtain the corresponding hidden vectors, and then the hidden vectors are aggregated to obtain the second eigenvector.

[0114] Specifically:

[0115] Construct a spatial adjacency matrix A from an undirected spatial subgraph S ; Construct the time adjacency matrix A based on the undirected time subgraph T ;

[0116] The spatial adjacency matrix A S , spatial node attribute matrix X S , spatial weight matrix W S Input to the first encoder to obtain the spatial hidden vector H S :

[0117]

[0118] Among them, σ represents the activation function, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0119] The time adjacency matrix A T , time node attribute matrix X T , time weight matrix W T Input to the second encoder to obtain the time hidden vector H T ;

[0120]

[0121] in, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained; the first encoder and the second encoder can adopt a graph convolutional neural network.

[0122] H S and the time hidden vector H T Input to the aggregation layer to obtain the second eigenvector H (2) :H (2) =aggregate(H S ,H T ), where aggregate(.) represents the aggregation function.

[0123] The aggregation function can be selected as a weighted function, that is, the pre-set spatial hidden vector H S and the time hidden vector H T The weights of the hidden vector H are then summed up to obtain the weights of the hidden vector H. S and the time hidden vector H T Perform weighted summation to obtain the final second eigenvector H (2) .

[0124] The preprocessed environmental information is input into the first neural network for encoding to obtain the first eigenvector H (1) In the present invention, the first neural network adopts a bidirectional long short-term memory network BI-LSTM.

[0125] After obtaining the first eigenvector H (1) and the second eigenvector H (2) Finally, the first eigenvector and the second eigenvector are concatenated to obtain the third eigenvector: H (3) =concat(H (1) ,H (2) ).

[0126] The third eigenvector is input into the recurrent neural network model to output the abnormal probability distribution of the monitoring node: P = RNN (H (3) ).

[0127] The recurrent neural network model includes a recurrent neural network layer, a pooling layer and a softmax layer. The recurrent network layer is used to extract features from the third eigenvector to obtain key features of each node in the third eigenvector.

[0128] The pooling layer is used to perform pooling operations on key features;

[0129] The Softmax layer is used to output the probability of abnormality of each monitoring node and obtain the abnormal probability distribution;

[0130] Where W 3 is the classifier parameter, b is the bias term, is the third eigenvector H (3) The output values ​​of the recurrent network layer and the pooling layer. For example, for the three monitoring nodes A, B, and C, the probability of abnormality of monitoring node A is 0.85, the probability of abnormality of monitoring node B is 0.45, and the probability of abnormality of monitoring node C is 0.2, thus obtaining the abnormal probability distribution (0.85, 0.45, 0.95).

[0131] S4: Based on the dam safety monitoring visualization system, the abnormal information of the dam is displayed.

[0132] After obtaining the abnormal probability distribution of the monitoring nodes, the monitoring nodes whose probability exceeds the preset threshold are determined as abnormal detection nodes and displayed through the dam safety monitoring visualization system. Specifically, in the digital dam structure model, the abnormal monitoring nodes are displayed in the second display mode, and the normal monitoring nodes are displayed in the first display mode, which is different from the first display mode; click on the monitoring node to view the monitoring data table and / or result map associated with the node in the form of a pop-up window.

[0133] In order to further improve the timeliness and operation and maintenance capabilities of dam safety monitoring and improve user experience, the present invention also constructs a dam safety monitoring visualization system, such as Figure 3 As shown in the figure, the system includes an operation tree, a central window and an auxiliary window, which can realize the visualization of the dam engineering structure, the monitoring design (monitoring items, monitoring points, monitoring instruments), the monitoring data and the monitoring alarm status, including:

[0134] Use BIM technology to build a digital dam structure model, and use BIM+GIS technology to visualize the spatial distribution of the digital dam structure model based on survey and design data;

[0135] Bind the locations of benchmark points and monitoring points to the spatial coordinates of the digital dam structure model, distinguish benchmark points and monitoring points with different icons, and manage the spatial distribution of benchmark points, monitoring points and monitoring items in the form of layers;

[0136] Visualize the topology of monitoring instruments (RTU, sensors, etc.) and associate the working status information of the monitoring instruments. Monitor instruments are displayed with different symbols, and the working status of the instruments is displayed in red and green on the topology diagram. Red indicates instrument failure and green indicates normal operation.

[0137] Bind the sensors to the corresponding monitoring points and manage the spatial distribution of the sensors in the form of layers; bind the RTU positions to the spatial coordinates of the digital dam structure model and associate the relevant sensors to manage the spatial distribution of the RTU in the form of layers; bind the monitoring data to the corresponding monitoring points and click on the monitoring points to view the corresponding monitoring data tables and results maps in the form of pop-up windows;

[0138] Set different display forms of monitoring nodes to realize the visualization of alarm status. When it is judged as an abnormal monitoring node, the corresponding monitoring point icon is displayed in the form of a flashing red light. When it is judged as a normal monitoring node, the corresponding monitoring point icon remains in a green light state.

[0139] Click the abnormal monitoring point icon to view the monitoring data table and results chart, analyze the cause of the monitoring anomaly, and quickly develop a disposal plan.

[0140] Reference Figure 4 The present invention also provides a dam safety monitoring device, which comprises:

[0141] A collection module is used to collect multi-source monitoring time series data, wherein the multi-source monitoring time series data includes dam structure status information and environmental information; wherein the dam structure status information includes stress, strain, deformation, seepage pressure and flow; and the environmental information includes one or more of reservoir water level, air temperature, air pressure, rainfall and evaporation;

[0142] A preprocessing module, used for preprocessing the multi-source monitoring time series data to obtain the multi-source monitoring time series data to be processed; the multi-source monitoring time series data to be processed includes the dam structure state information to be processed and the environmental information to be processed;

[0143] An anomaly detection module is used to input the multi-source monitoring time series data to be processed into a trained anomaly detection model to identify and obtain dam anomaly information; the anomaly detection model includes a first deep neural network, a graph neural network, and a second deep neural network; the multi-source monitoring time series data to be processed is encoded using the first deep neural network and the graph neural network to obtain a third feature vector, and the third feature vector is decoded using the second deep neural network to obtain the dam anomaly information;

[0144] The visualization module is used to display the abnormal information of the dam based on the dam safety monitoring visualization system.

[0145] The anomaly detection module uses the first deep neural network and the graph neural network to encode the processed data to obtain the third feature vector, which specifically includes:

[0146] An undirected graph is constructed by utilizing the propagation characteristics of the dam structure status information and multiple monitoring nodes;

[0147] Inputting environmental information into a first deep neural network to obtain a first feature vector;

[0148] Input the undirected graph into the graph neural network to obtain the second eigenvector;

[0149] The first eigenvector and the second eigenvector are fused to obtain a third eigenvector.

[0150] The anomaly detection module uses the second deep neural network to decode the third feature vector to obtain dam anomaly information, specifically including: inputting the third feature vector into the second deep neural network to obtain the distribution of abnormal monitoring nodes.

[0151] Using the propagation characteristics of the dam structure status information and multiple monitoring nodes, an undirected graph is constructed, including:

[0152] The stress, strain, deformation, seepage pressure and flow of all monitoring nodes are obtained by using stress propagation model, strain propagation model, deformation propagation model, seepage pressure propagation model and flow propagation model;

[0153] Get all monitoring nodes at time t, connect adjacent nodes by edge according to spatial position, and construct an undirected spatial subgraph. The distance between monitoring nodes in the undirected spatial subgraph is the edge weight W. S , the stress, strain, deformation, seepage pressure and flow of the monitoring node are used as the spatial node attribute matrix X S ;

[0154] Get all monitoring nodes at time t-t1, t, and t+t2, connect adjacent nodes by edges according to the time distribution, and construct an undirected time subgraph, where the time step length in the undirected time subgraph is the edge weight WT The stress, strain, deformation, seepage pressure and flow of the node are used as the time node attribute matrix X T .

[0155] Input the undirected graph into the graph neural network to obtain the second eigenvector, which specifically includes:

[0156] Construct the spatial adjacency matrix A S and the time adjacency matrix A T ;

[0157] The spatial adjacency matrix A S , spatial node attribute matrix X S , spatial weight matrix W S Input to the first encoder to obtain the spatial hidden vector H S :

[0158]

[0159] Among them, σ represents the activation function, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0160] The time adjacency matrix A T , time node attribute matrix X T , time weight matrix W T Input to the second encoder to obtain the time hidden vector H T ;

[0161]

[0162] in, I represents the identity matrix, Indicates based on The corresponding degree matrix obtained;

[0163] H S and the time hidden vector H T Input to the aggregation layer to obtain the second eigenvector H (2) :H (2) =aggregate(H S ,H T ), where aggregate(.) represents the aggregation function.

[0164] Inputting the environmental information into the first deep neural network to obtain a first feature vector specifically includes: inputting the environmental information into a bidirectional long short-term memory neural network for encoding to obtain a first feature vector H (1) .

[0165] The first eigenvector and the second eigenvector are merged to obtain the third eigenvector, which specifically includes: concatenating the first eigenvector and the second eigenvector to obtain the third eigenvector: H (3) =concat(H (1) ,H (2) ).

[0166] Input the third feature vector into the second deep neural network to obtain the distribution of abnormal monitoring nodes, specifically including: input the third feature vector into the recurrent neural network model, and output the probability distribution of abnormal monitoring nodes: P = RNN (H (3) ).

[0167] The monitoring nodes whose probability exceeds the preset threshold are regarded as abnormal monitoring nodes.

[0168] The visualization module displays abnormal information based on the dam safety monitoring visualization system, including:

[0169] The monitoring nodes whose probability exceeds the preset threshold are regarded as abnormal monitoring nodes;

[0170] In the digital dam structure model, abnormal monitoring nodes are displayed in a second display mode, and normal monitoring nodes are displayed in a first display mode, wherein the second display mode is different from the first display mode;

[0171] Click a monitoring node to view the monitoring data table and / or achievement graph associated with the node in a pop-up window.

[0172] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0173] The present invention also provides a computer-readable storage medium for storing computer instructions, wherein the computer instructions implement the steps of the aforementioned method when executed by a processor.

[0174] The dam safety monitoring method, device, computer equipment and storage medium provided by the present invention have the following advantages:

[0175] (1) When performing anomaly detection on dam monitoring data, the present invention constructs an anomaly detection model including a first deep neural network, a graph neural network, and a second deep neural network. The environmental characteristics of the environment in which the dam is located are extracted by the first deep neural network, and the inherent characteristics of the dam structure state information are extracted by the graph neural network. The environmental characteristics and the inherent characteristics are fused and input into the second deep neural network, and the dam anomaly information is decoded to obtain the dam anomaly information. Since the environmental characteristics and the inherent characteristics are comprehensively considered, the accuracy of dam anomaly detection is improved.

[0176] (2) When obtaining the inherent characteristics of the dam structure status information, the present invention first uses the propagation model to obtain the stress, strain, deformation, seepage pressure and flow value of all monitoring nodes, and uses the value as the attribute of the node in the undirected graph, providing a solid data foundation for dam anomaly detection; at the same time, the present invention constructs an undirected spatial subgraph based on the relative positions of all monitoring nodes at the same time, and constructs a spatial weight matrix according to the distance between the monitoring nodes; constructs an undirected time subgraph based on the monitoring nodes at different times, constructs a time weight matrix according to the time step, and uses the stress, strain, deformation, seepage pressure and flow value of the monitoring node as the attribute of the monitoring node, and then encodes the undirected subgraph through the graph neural network to obtain the inherent characteristics of the dam status information. Since the distance characteristics between the monitoring nodes, the time series characteristics of the monitoring data, and the propagation characteristics of the monitoring data are taken into account, the inherent characteristics obtained by encoding are more in line with the monitoring node, further improving the accuracy of dam anomaly detection.

[0177] (3) When the present invention uses a graph neural network to encode an undirected graph, the hidden vector of the spatial dimension is obtained by the first encoder, and the hidden vector of the time dimension is obtained by the second encoder. Then, the hidden vector of the spatial dimension and the hidden vector of the time dimension are aggregated to obtain the feature vector of the monitoring node. Since the hidden vectors of the time dimension and the spatial dimension are aggregated, the inherent characteristics obtained by the encoding have spatial characteristics and temporal characteristics, which once again improves the accuracy of dam anomaly detection.

[0178] (4) The present invention constructs an integrated visualization system for associating and displaying abnormal monitoring points, abnormal information and monitoring data in a digital dam spatial structure model, thereby improving the timeliness and operation and maintenance capabilities of dam safety monitoring and improving user experience.

[0179] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density digital video disc (DVD)), or a semiconductor medium (eg, a solid state disc (SSD)).

[0180] The memory in the embodiment of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read only memory (ROM), a programmable read only memory (PROM), an erasable programmable read only memory (EPROM), an electrically erasable programmable read only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM). It should be noted that the memory of the method described in the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0181] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in conjunction with the embodiment of the present application can be directly embodied as a hardware processor for execution, or a combination of hardware and software modules in a processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it is not described in detail here.

[0182] It should be noted that the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor can be combined and performed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0183] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of application involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned application concept. For example, the above features are replaced with (but not limited to) technical features with similar functions applied in the present application.

Claims

1. A dam safety monitoring method, characterized in that: The following steps are involved: S1: Collect multi-source monitoring time series data, wherein the multi-source monitoring time series data includes dam structure status information and environmental information; wherein the dam structure status information includes stress, strain, deformation, seepage pressure and flow; and the environmental information includes one or more of reservoir water level, temperature, air pressure, rainfall and evaporation; S2: preprocessing the multi-source monitoring time series data to obtain the multi-source monitoring time series data to be processed; the multi-source monitoring time series data to be processed includes the dam structure state information to be processed and the environmental information to be processed; S3: Input the multi-source monitoring time series data to be processed into the trained anomaly detection model to identify and obtain dam anomaly information, including: The anomaly detection model includes a first deep neural network, a graph neural network, and a second deep neural network; Encoding the to-be-processed multi-source monitoring time series data using the first deep neural network and the graph neural network to obtain a third feature vector; Decoding the third feature vector using the second deep neural network to obtain dam abnormality information; S4: Based on the dam safety monitoring visualization system, the abnormal information of the dam is displayed.

2. A dam safety monitoring method according to claim 1, characterized in that: The using the first deep neural network and the graph neural network to encode the multi-source monitoring time series data to be processed to obtain a third feature vector specifically includes: An undirected graph is constructed using the propagation model of the dam structure status information and the location information of the monitoring nodes; Inputting the environmental information to be processed into a first deep neural network to obtain a first feature vector; Input the undirected graph into the graph neural network to obtain the second eigenvector; Merging the first eigenvector and the second eigenvector to obtain a third eigenvector; The using the second deep neural network to decode the third feature vector to obtain dam abnormality information specifically includes: The third feature vector is input into the second deep neural network to obtain the abnormal probability distribution of the monitoring node.

3. A dam safety monitoring method according to claim 2, characterized in that: The use of the propagation model of the dam structure status information and the monitoring node location information to construct an undirected graph specifically includes: Using a stress propagation model, a strain propagation model, a deformation propagation model, a seepage pressure propagation model, and a flow propagation model, the properties of all monitoring nodes are obtained, wherein the properties of the monitoring nodes include stress, strain, deformation, seepage pressure, and flow; Obtain the attributes of all monitoring nodes at time t, connect adjacent nodes by edge according to the spatial position of the monitoring nodes, and construct an undirected spatial subgraph; construct a spatial weight matrix W according to the distance between the monitoring nodes in the undirected spatial subgraph S , construct the spatial node attribute matrix X according to the stress, strain, deformation, seepage pressure and flow of the monitoring node S ; Obtain the attributes of all monitoring nodes at time t-t1, t, and t+t2, connect adjacent nodes by edges according to the time distribution, and construct an undirected time subgraph; construct a time weight matrix W according to the time step in the undirected time subgraph T , construct the time node attribute matrix X according to the stress, strain, deformation, seepage pressure and flow of the monitoring node T .

4. A dam safety monitoring method according to claim 3, characterized in that: The step of inputting the undirected graph into the graph neural network to obtain the second feature vector specifically includes: The graph neural network includes a first encoder, a second encoder and an aggregation layer; Construct a spatial adjacency matrix A from an undirected spatial subgraph S ; Construct the temporal adjacency matrix A from the undirected temporal subgraph T ; The spatial adjacency matrix A S , spatial node attribute matrix X S , spatial weight matrix W S Input to the first encoder to obtain the spatial hidden vector H S : Among them, σ represents the activation function, I represents the identity matrix, According to The corresponding degree matrix obtained; The time adjacency matrix A T , time node attribute matrix X T , time weight matrix W T Input to the second encoder to obtain the time hidden vector H T ; in, I represents the identity matrix, According to The corresponding degree matrix obtained; H S and the time hidden vector H T Input to the aggregation layer to obtain the second eigenvector H (2) :H (2) =aggregae(H S , H T ), where aggregate(.) represents the aggregation function.

5. A dam safety monitoring method according to claim 4, characterized in that: The step of inputting the environmental information to be processed into a first deep neural network to obtain a first feature vector; and fusing the first feature vector and the second feature vector to obtain a third feature vector specifically includes: The first deep neural network adopts a bidirectional long short-term memory neural network; the environmental information to be processed is input into the bidirectional long short-term memory neural network for encoding to obtain a first feature vector H (1) ; Use the following formula to convert the first eigenvector H (1) and the second eigenvector H (2) Splice and get the third eigenvector H (3) :H (3) =concat(H (1) , H (2) ).

6. A dam safety monitoring method according to claim 5, characterized in that: Inputting the third feature vector into the second deep neural network to obtain the abnormal probability distribution of the monitoring node specifically includes: The second deep neural network adopts a recurrent neural network model; The third eigenvector H (3) Input to the recurrent neural network model and output the abnormal probability distribution P of the monitoring node: P = RNN (H (3) ).

7. A dam safety monitoring method according to claim 2, characterized in that: Also includes: Constructing a dam safety monitoring visualization system, wherein the dam safety monitoring visualization system includes an operation tree, a central window, and an auxiliary window; The dam safety monitoring visualization system is used to display dam abnormality information, specifically including: The monitoring nodes whose probability exceeds the preset threshold are regarded as abnormal monitoring nodes; In the digital dam structure model, abnormal monitoring nodes are displayed in a second display mode, and normal monitoring nodes are displayed in a first display mode; the second display mode is different from the first display mode; In the digital dam structure model, click on a monitoring node to view the monitoring data table and / or results diagram associated with the monitoring node in a pop-up window.

8. A dam safety monitoring device, characterized in that: Includes the following modules: A collection module is used to collect multi-source monitoring time series data, wherein the multi-source monitoring time series data includes dam structure status information and environmental information; wherein the dam structure status information includes stress, strain, deformation, seepage pressure and flow; and the environmental information includes one or more of reservoir water level, air temperature, air pressure, rainfall and evaporation; A preprocessing module, used for preprocessing the multi-source monitoring time series data to obtain the multi-source monitoring time series data to be processed; the multi-source monitoring time series data to be processed includes the dam structure state information to be processed and the environmental information to be processed; An anomaly detection module is used to input the multi-source monitoring time series data to be processed into a trained anomaly detection model to identify and obtain dam anomaly information; the anomaly detection model includes a first deep neural network, a graph neural network, and a second deep neural network; the multi-source monitoring time series data to be processed is encoded using the first deep neural network and the graph neural network to obtain a third feature vector, and the third feature vector is decoded using the second deep neural network to obtain the dam anomaly information; The visualization module is used to display the abnormal information of the dam based on the dam safety monitoring visualization system.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images