A dam seepage field reconstruction and leakage channel inversion method and device

Abstract

The application discloses a dam seepage field reconstruction and leakage channel inversion method and equipment, and the method comprises the steps of: constructing a structured topological graph reflecting the adjacency relationship between measuring points according to the spatial distribution characteristics of the dam seepage pressure monitoring section, and converting the monitoring data into a multi-dimensional feature tensor containing seepage pressure values, change rates and environmental factors; constructing a topological constraint-based space-time graph attention network model, introducing a topological adjacency mask in the attention score calculation process of the model graph attention layer to force the model to only aggregate features between nodes that exist in the topological structure, and embedding the spatial geometric constraint into the deep learning weight updating process; analyzing the attention weight matrix obtained by model training to invert the dominant leakage path inside the dam body, and outputting the reconstructed three-dimensional seepage field distribution based on sparse measuring points. The application eliminates the pseudo-correlation interference in the data-driven model through the topological hard constraint mechanism, and can realize the diagnosis of the dam seepage state.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and diagnosis technology for the safety of water conservancy and hydropower projects. Specifically, it relates to a method, equipment, and storage medium that combines graph theory topology and deep learning attention mechanism to perform high-precision reconstruction of the three-dimensional seepage field inside the dam and inversion of hidden leakage channels based on topological constraints using sparse seepage pressure monitoring data. Background Technology

[0002] As a key infrastructure for intercepting rivers and regulating water resources, the structural safety of dams is directly related to the safety of people's lives and property downstream.

[0003] Seepage is the primary factor affecting dam safety. According to statistics, a considerable proportion of earth-rock dam failures are caused by seepage damage such as piping and soil erosion. Therefore, accurately grasping the distribution of the seepage field inside the dam body and timely identifying potential concentrated seepage channels are the core prerequisites for achieving early warning and precise reinforcement of earth-rock dam safety.

[0004] Existing seepage monitoring and analysis methods are mainly divided into two categories: traditional physical causal models and statistical models. The physical genesis model is based on Darcy's law and the seepage continuity equation. It uses the finite element method for numerical calculation and has a clear physical meaning. However, it requires precise boundary conditions and seepage parameters of the soil and rock mass. In actual engineering, the material parameters inside the dam body often have significant spatial variability and are difficult to obtain, resulting in a large deviation between the calculation results and the measured values. Statistical models focus on establishing mathematical correlations between environmental quantities such as water level and rainfall and seepage pressure effects. Although the calculations are simple, they lack consideration for spatial physical structures, only achieve single-point fitting, cannot describe the spatial hydraulic connections between measuring points, and cannot reconstruct the seepage field distribution in areas where sensors are not deployed.

[0005] In recent years, data-driven methods, represented by deep learning, have been gradually applied to dam safety monitoring. However, existing general neural network models still have significant technical limitations in practical engineering applications: First, traditional models usually treat piezometers distributed in different spatial locations as independent variables, ignoring the physical topological connections of the dam structure. This leads to the model easily learning spurious correlations between nodes that are spatially far apart and have weak hydraulic connections, violating basic seepage physics. Second, deep learning models generally have a "black box" characteristic; although they can provide predictions, they cannot explain the physical logic behind the predictions. In addition, due to monitoring costs, the deployment of dam piezometers is usually sparse. How to accurately reconstruct the complete three-dimensional seepage field based on limited sparse measurement point information has always been a technical bottleneck restricting the accurate location of dam hazards.

[0006] To address this issue, this application proposes a method for reconstructing the seepage field and inverting seepage channels in dams. This method effectively integrates the physical topology of the dam and possesses both physical interpretability and seepage channel inversion capabilities, thereby resolving the aforementioned technical problems. Summary of the Invention

[0007] The main objective of this invention is to provide a method for reconstructing the seepage field and inverting seepage channels in dams. This method aims to overcome the technical shortcomings of existing monitoring models, such as ignoring spatial topology, lacking physical interpretability, and having poor reconstruction capabilities for sparse measuring points. It achieves high-precision reconstruction of the three-dimensional seepage field inside the dam and physical inversion of potential seepage channels, thereby solving the technical problems mentioned in the background art.

[0008] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: A method for reconstructing the seepage field and inverting seepage channels in a dam, executed by computer equipment, includes the following steps: Step S1. Topology Construction and Feature Engineering: Based on the spatial distribution characteristics of the dam seepage pressure monitoring sections, a structured topology map reflecting the adjacency relationship between monitoring points is constructed, and the monitoring data is transformed into a multidimensional feature tensor containing seepage pressure values, change rates, and environmental factors. Step S2. Physical constraint model training: A spatiotemporal graph attention network model based on topological constraints is constructed based on structured topological graphs and multidimensional feature tensors. A topological adjacency mask is introduced in the attention score calculation process of the model graph attention layer to force the model to perform feature aggregation only between nodes that are connected in the topological structure, and to embed spatial geometric constraints into the deep learning weight update process. Step S3. Reconstruction and Inversion Application: The attention weight matrix obtained from the analytical model training is used to invert the dominant seepage path inside the dam body, and the three-dimensional seepage field distribution is reconstructed based on the output of sparse measuring points.

[0009] Preferably, the structured topology graph construction process in step S1 includes: Obtain the spatial coordinates and historical monitoring data of each piezometer on the dam; Based on the cross-section and elevation distribution of the measuring points, the dam monitoring cross-section is abstracted into a two-dimensional grid coordinate system. For any measuring point... Search for its neighboring nodes in the grid coordinate system. A topological adjacency matrix reflecting the spatial connection relationship between measurement points is established according to the preset adjacency rules.

[0010] Preferably, the adjacency rules for constructing the structured topology graph in step S1 include: If node If one of the following conditions is met, then in the topological adjacency matrix... Setting Mark as having a connection relationship; otherwise set Marked as no connection: Condition 1: Node With nodes Located at the same elevation and in adjacent cross sections; Condition 2: Node With nodes Located on the same cross section and at adjacent elevations; Condition 3: Node With nodes Located at adjacent cross-sections and adjacent elevations; This rule ensures that the graph structure accurately reflects the physical path of fluid diffusion in porous media along horizontal, vertical, and diagonal directions.

[0011] Preferably, the specific calculation process of introducing the topological adjacency mask in the attention score calculation process in step S2 is as follows: After calculating the original attention scores between nodes, a mask matrix corresponding to the shape of the topological adjacency matrix is ​​superimposed, resulting in:

[0012] in, Represents a node For nodes Attention weights The original attention coefficients are calculated from the dot product of the query vector and the key vector; For positions marked as unconnected in the topological adjacency matrix, the corresponding element in the mask matrix is ​​set to negative infinity. For positions marked as connected, the corresponding element in the mask matrix is ​​set to zero. The topological adjacency mask is defined as follows:

[0013] in, The elements of the topological adjacency matrix constructed in step S1; when Attention weights after softmax function operation Attention weight approaches zero The expanded calculation form is as follows:

[0014] in, This represents the total number of dam seepage pressure monitoring nodes. This is the index variable for traversing all nodes; In the subsequent normalization calculation step, the mask value of negative infinity will make the attention weights between non-adjacent nodes approach zero, thereby realizing the hard constraint of physical mechanism on the weights of neural network.

[0015] Preferably, the specific process of the spatial geometric constraint embedding deep learning weight update process in step S2 includes: By forcibly resetting the attention weights between non-topologically adjacent nodes to zero through masking operations, the model is trained using the multidimensional feature tensor to learn the spatiotemporal correlation features constrained by the topological structure.

[0016] Preferably, the method for constructing the multidimensional feature tensor includes: Calculate the first-order time difference of each piezometer to obtain Characteristics of the rate of change of osmotic pressure at time t time ; Reservoir water level monitoring data Perform spatial dimension broadcasting to match the number of nodes; The original seepage pressure values ​​of each node , and after the broadcast Normalization is performed and the data is concatenated along the feature dimensions to generate a shape of... The input tensor, where The time step of the input sequence. 3 represents the total number of dam seepage pressure monitoring nodes, and 3 represents the number of characteristic channels.

[0017] Preferably, the process for inverting the dominant seepage path inside the dam body in step S3 includes: Obtain and calculate the mean of the attention weight matrix obtained during model training over the time dimension. , as the global correlation strength matrix; Calculate the absolute value of the rate of change of seepage pressure at each measuring point at the predicted time. As a node activity vector , used to characterize the local seepage activity; Define connecting edges Leakage risk index for: , For connecting edges The mean of the attention weight matrix over the time dimension. For nodes The activity vector; Multiplying the global correlation strength by the local seepage activity yields the leakage risk index for each connection edge. Screening out leakage risk index Connection edges exceeding a preset threshold serve as the dominant leakage channels, and paths are constructed and connected in three-dimensional space. This method transforms abstract neural network weights into engineering-visible seepage paths, exhibiting strong physical interpretability.

[0018] Preferably, the spatiotemporal graph attention network model includes a physical constraint space layer and a temporal evolution layer, wherein: A physical constraint space layer is used to extract spatial features according to the calculation formula; The time evolution layer employs a gated recurrent unit network to take the feature sequence output from the physical constraint space layer as input and output the predicted seepage pressure value for the next time step.

[0019] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0020] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0021] As can be seen from the above technical solution, the present invention provides a method for reconstructing the seepage field and inverting seepage channels in a dam. Compared with the prior art, the present invention has the following advantages: 1. This invention uses topological adjacency masking for the attention mechanism, which can hard-encode the physical spatial structure of the dam into the deep learning model, forcing the model to follow the physical connection path for information transmission. This effectively avoids spurious correlation predictions that violate physical common sense generated by purely data-driven models, and its physical mechanism constraint is strong.

[0022] 2. This invention integrates structured mapping with multidimensional hydraulic features, enabling the model to effectively capture the nonlinear spatiotemporal dependencies between sparse measuring points. Combined with interpolation algorithms, it can further reconstruct the three-dimensional seepage field distribution with only a small number of measuring points, reducing the dependence on high-density monitoring networks, and ultimately achieving high accuracy in sparse data reconstruction.

[0023] 3. In summary, this invention eliminates spurious correlation interference in the data-driven model through a topological hard constraint mechanism, enabling the diagnosis of dam seepage behavior.

[0024] It should be understood that the descriptions in this section are not intended to identify key or essential features of embodiments of the invention, nor are they intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Of course, implementing any product of the invention does not necessarily require achieving all of the advantages described above simultaneously. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a planar schematic diagram of the structured topology mapping rules for dam monitoring points in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the graph attention calculation mechanism that incorporates topological adjacency masks in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] For details in the embodiments, please refer to Figures 1 to 3 .

[0028] like Figure 1 As shown, the dam seepage field reconstruction and leakage channel inversion method proposed in this embodiment of the invention eliminates spurious correlation interference in the data-driven model through a topological hard constraint mechanism, enabling diagnosis of dam seepage behavior. The method mainly includes the following steps: Step S1: Obtain the spatial coordinates and historical monitoring data of each piezometer of the dam. Based on the cross section and elevation distribution of the measuring points, establish a topological adjacency matrix that reflects the spatial connection relationship between the measuring points according to the preset adjacency rules, and integrate the piezometer data and environmental monitoring data to construct a multidimensional feature tensor. S11. Data Acquisition and Preprocessing Acquire basic data from the dam seepage monitoring system, including Historical reading sequence of piezometer points and the corresponding reservoir water level sequence ,in Indicates a time step. This indicates the measurement point number. The raw data is cleaned to remove outliers caused by equipment malfunctions, and missing data is filled using linear interpolation. Subsequently, all data are resampled to a uniform time interval. (1 hour).

[0029] S12. Establish a structured topological adjacency matrix Based on the layout of the dam monitoring sections, a two-dimensional grid coordinate system is established, such as... Figure 2 As shown. Set the measuring point. The section number is Its hole number in the elevation direction is Define the topological adjacency matrix of the dam seepage monitoring network. For any two measuring points and Determine the spatial connectivity of the nodes. A node is considered a node if one of the following conditions is met. With nodes There is a physical connection, so Otherwise, let The connection determination conditions are as follows: (1) Horizontally adjacent: and ; (2) Vertically adjacent: and ; (3) Diagonally adjacent: and .

[0030] The above rules construct a sparse graph structure with horizontal, vertical and diagonal connections, which can simulate the diffusion path of seepage in the dam medium in various directions.

[0031] S13. Constructing a multidimensional hydraulic characteristic tensor To enhance the model's ability to perceive physical laws, this embodiment constructs an input tensor containing three feature channels. For each time step and each measuring point Extract the following features: (1) Normalized osmotic pressure :

[0032] in, and These are the mean and standard deviation of historical osmotic pressure data, respectively. For measuring points At time step The original osmotic pressure value.

[0033] (2) Normalized rate of change of osmotic pressure : First, calculate the first-order time difference. This indicator reflects the local seepage velocity; it is then normalized.

[0034] in, and These are the population mean and population standard deviation of the first-order time difference data of seepage pressure at all measuring points, respectively.

[0035] (3) Normalized environmental driving factors : Normalize the reservoir water level data:

[0036] in, For measuring points At any moment The original reservoir water level data, and These represent the mean and standard deviation of historical reservoir water level data, respectively.

[0037] Since the reservoir water level is a global variable affecting the entire dam, broadcasting it spatially allows each measuring point to... At any moment The third-dimensional features are all .

[0038] Finally, the three features mentioned above are concatenated to obtain the time step. Feature matrix And according to the length of the sliding window Constructing the temporal tensor of the model input ,in For batch size, The time step length of the input sequence. 3 represents the total number of monitoring points at the dam, and 3 represents the number of characteristic channels. It is the field of real numbers.

[0039] Step S2: Construct a spatiotemporal graph attention network model. In the attention score calculation process of the graph attention layer, a topological adjacency mask is introduced. This masking operation forces the attention weights between non-topologically adjacent nodes to be reset to zero. The model is then trained using a multidimensional feature tensor to learn spatiotemporal correlation features constrained by the topological structure, such as... Figure 3 As shown; S21. Construct a physical constraint space layer This layer aims to extract spatially correlated features constrained by the dam's physical topology. For each time step in the input feature sequence... It employs a multi-head graph attention mechanism.

[0040] First, input feature vector Mapped to a query vector using a learnable linear transformation matrix. Key vector Sum value vector ,have:

[0041]

[0042]

[0043] Among them, superscript Indicates the first One point of attention; For measuring points At time step The three-dimensional input feature vector; The first The query, key, and value weight matrices corresponding to each attention point are all learnable parameters that are continuously updated during model training.

[0044] S22. Introducing attention calculation of topological adjacency mask. To implement hard constraints of physical topology on neural networks, this embodiment introduces a topological adjacency mask when calculating attention scores.

[0045] First, compute nodes With nodes raw attention scores between ,have:

[0046] in, To query the feature dimensions of the vector and key vector, Used to scale the dot product result to prevent gradient vanishing; Indicates the transpose of the query vector; Next, define the topological adjacency mask matrix. Its elements The adjacency matrix obtained from step S12 Decide:

[0047] Then, the mask matrix is ​​superimposed on the original attention score, and the normalized attention weights are calculated using the Softmax function. :

[0048] in, To traverse all of the dam Index variables for each measurement point; For the natural constant An exponential function with base 0; because This formula forces the attention weight between non-physically adjacent nodes to be zero, thereby blocking the transmission of invalid spatial information.

[0049] Finally, the value vectors are weighted and aggregated according to the attention weights to obtain the spatial feature output. :

[0050] in, This refers to the total number of attention heads in a multi-head attention mechanism. This means concatenating the outputs of all attention heads along the feature dimension. To output the linear mapping weight matrix.

[0051] S23. Temporal Evolution and Decoding To capture the dynamic hysteresis effect of the seepage process, the spatial feature sequence output by S22 is... The input is fed into a gated recurrent unit (GRU) network for timing modeling:

[0052] in, For GRU networks at time steps Extracted hidden layer state feature vectors, This represents the hidden state of the previous time step. This is the nonlinear calculation function for the gated cyclic unit.

[0053] Take the hidden state of the last time step of the GRU network The predicted seepage pressure value at future moments is obtained through mapping using a fully connected decoding layer. :

[0054] in, This represents the linear mapping operation of a fully connected layer. To include all The vector of predicted seepage pressure values ​​at each measuring point at the target time.

[0055] S24. Model Training and Optimization Mean squared error (MSE) is used as the loss function. Measure the predicted osmotic pressure value Compared with the actual monitoring value Differences between them:

[0056] in, BatchSize is the number of samples in a single training iteration. This represents the total number of measurement points. The first in the batch The first sample Predicted seepage pressure values ​​at each measuring point; This corresponds to the actual monitored seepage pressure value.

[0057] Using the Adam optimization algorithm, the loss function is minimized based on the backpropagation mechanism. Iteratively update the weight parameters in the model And the GRU network parameters, until the model's error on the validation set converges, to obtain the trained physical perception model.

[0058] By employing a topological adjacency mask for the attention mechanism, the physical spatial structure of the dam can be hard-coded into the deep learning model, forcing the model to follow physical connection paths for information transmission. This effectively avoids spurious correlation predictions that violate physical common sense generated by purely data-driven models, and its physical mechanism constraints are strong.

[0059] Step S3: Use the trained model to predict the seepage pressure values ​​of each measuring point at future times, reconstruct the three-dimensional seepage field distribution through interpolation algorithm, extract the attention weight matrix in the model, and combine the local seepage activity of each node to invert the dominant seepage channel path inside the dam.

[0060] S31. Three-dimensional seepage field interpolation reconstruction based on sparse measuring points Using the model trained in step S2, input the latest monitoring data sequence, and output the predicted seepage pressure values ​​for all measuring points at future times. Since the piezometers are sparsely distributed discrete points within the dam body, this embodiment employs an interpolation algorithm for field reconstruction in order to obtain a continuous seepage field distribution within the dam body.

[0061] Set up measuring points The physical coordinates are The predicted value is Construct a refined, rule-based grid covering the monitoring area. The cubic spline interpolation method is used to calculate the grid points. Estimated seepage pressure at the location :

[0062] in, The coordinates of any target interpolation point in the generated refined mesh; This represents a spatial interpolation function.

[0063] Based on the generated gridded data Drawing a three-dimensional seepage field surface diagram or equipotential line cloud map of the dam profile can intuitively display the water head distribution gradient inside the dam body, helping engineers to identify abnormally high water head areas or abnormal uplift of the phreatic line in areas where no sensors are deployed.

[0064] S32. Dominant Leakage Channel Inversion Based on Attention Weights This step transforms the "attention weights" in the deep learning model into the "hydraulic connection strength" in hydraulics, enabling the visualization and inversion of the leakage channel.

[0065] First, extract the attention weight matrix output by the physical constraint space layer of the model at the prediction time. If the model uses a multi-head attention mechanism, then the average weight of each head is taken:

[0066] in, The total number of heads in the multi-head attention mechanism; For the first Nodes in each attention head For nodes The weights; Represents a node For nodes The combined influence intensity physically reflects the seepage connectivity between the two.

[0067] Secondly, calculate the absolute value of the rate of change of seepage pressure at each measuring point at the predicted time. (Refer to step S13) This index approximately characterizes the local seepage velocity or activity at that location.

[0068] Define connecting edges Leakage risk index :

[0069] The physical meaning of this formula is: only when the upstream node The seepage activity is high (fast flow rate), and the nodes With nodes A path is considered a seepage channel only when the connectivity between them is strong (high weight).

[0070] Set risk threshold (85th percentile of all marginal risk indices), filter out those that meet the criteria The connecting edges. In the 3D space model, these high-risk edges are drawn as line segments with directional arrows, and the thickness of the line segments is... Proportional. The interconnected high-risk edges form the dominant seepage channel network inside the dam, indicating the main source and direction of seepage.

[0071] In summary, this method integrates structured mapping with multidimensional hydraulic features, enabling the model to effectively capture the nonlinear spatiotemporal dependencies between sparse measuring points. When combined with interpolation algorithms, it can further reconstruct the three-dimensional seepage field distribution with only a small number of measuring points, reducing the dependence on high-density monitoring networks, and ultimately achieving high accuracy in sparse data reconstruction.

[0072] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0073] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0074] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the dam seepage field reconstruction and leakage channel inversion methods in the above embodiments.

[0075] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above method.

[0076] This application also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other via the communication bus. Memory, used to store computer programs; The processor, when executing the program stored in memory, implements the above-mentioned method for reconstructing the dam seepage field and inverting seepage channels.

[0077] The communication bus mentioned in the above-mentioned electronic devices can be a standard bus for interconnecting peripheral components or an extended industrial standard structure bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.

[0078] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0079] The memory may include random access memory or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0080] The processors mentioned above can be general-purpose processors, including central processing units, network processors, etc.; they can also be digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0081] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0083] Furthermore, it should be noted that if any directional indication (such as up, down, left, right, front, back, etc.) is involved in the embodiments of the present invention, the directional indication is only used to explain the relative positional relationship and movement of each component in a specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0084] Furthermore, those skilled in the art should understand that in the actual use of the embodiments of this application, there may be preset thresholds used as the basis for judging the corresponding technical solutions. These thresholds are conventional technical means commonly used in the field to implement functions such as state judgment, condition recognition, and control logic switching. The specific values, setting basis, value selection methods, determination methods, and adjustment rules of the thresholds involved in this technical solution are all conventional technical choices that can be reasonably determined by those skilled in the art based on conventional technical factors such as actual application scenarios, system working states, characteristics of the detection object, hardware performance parameters, and functional requirements, through conventional experiments, calibrations, and debugging. The specific setting and adjustment of the aforementioned thresholds will not cause this technical solution to be unimplementable as a whole, nor will it affect the realization of the core concept and the achievement of the technical effects of this technical solution.

[0085] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the meaning of "and / or" throughout the text includes three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, in the embodiments of this invention, "multiple" refers to two or more. Moreover, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

Claims

1. A method for reconstructing the seepage field and inverting seepage channels in a dam, characterized in that, include: Based on the spatial distribution characteristics of the dam seepage pressure monitoring sections, a structured topology map reflecting the adjacency relationship between monitoring points is constructed, and the monitoring data is transformed into a multidimensional feature tensor containing seepage pressure values, change rates, and environmental factors. A spatiotemporal graph attention network model based on topological constraints is constructed based on structured topological graphs and multidimensional feature tensors. A topological adjacency mask is introduced in the attention score calculation process of the model graph attention layer to force the model to perform feature aggregation only between nodes that are connected in the topological structure, and to embed spatial geometric constraints into the deep learning weight update process. The attention weight matrix obtained from the analytical model training is used to invert the dominant seepage path inside the dam body, and the three-dimensional seepage field distribution is reconstructed based on the output of sparse measuring points. The process of constructing the structured topology graph includes: Obtain the spatial coordinates and historical monitoring data of each piezometer on the dam; Based on the cross-section and elevation distribution of the measuring points, the dam monitoring cross-section is abstracted into a two-dimensional grid coordinate system. For any measuring point... Search for its neighboring nodes in the grid coordinate system. A topological adjacency matrix reflecting the spatial connection relationship between measurement points is established according to the preset adjacency rules. The adjacency rules for constructing the structured topology graph include: If node If one of the following conditions is met, then in the topological adjacency matrix... Setting Otherwise set : Condition 1: Node With nodes Located at the same elevation and in adjacent cross sections; Condition 2: Node With nodes Located on the same cross section and at adjacent elevations; Condition 3: Node With nodes Located at adjacent cross-sections and adjacent elevations; The specific calculation process for the topological adjacency mask introduced in the attention score calculation is as follows: After calculating the original attention scores between nodes, a mask matrix corresponding to the shape of the topological adjacency matrix is ​​superimposed, resulting in: in, Represents a node For nodes Attention weights The original attention coefficients are calculated from the dot product of the query vector and the key vector; The topological adjacency mask is defined as follows: in, The elements of the topological adjacency matrix constructed in step S1; when Attention weights after softmax function operation Attention weight approaches zero The expanded calculation form is as follows: in, This represents the total number of dam seepage pressure monitoring nodes. This is the index variable for traversing all nodes; The specific process of embedding spatial geometric constraints into deep learning weight update includes: By forcibly resetting the attention weights between non-topologically adjacent nodes to zero through masking operations, the model is trained using the multidimensional feature tensor to learn the spatiotemporal correlation features constrained by the topological structure. The method for constructing the multidimensional feature tensor includes: Calculate the first-order time difference of each piezometer to obtain Osmotic pressure change rate characteristics at time intervals ; Reservoir water level monitoring data Perform spatial dimension broadcasting to match the number of nodes; The original seepage pressure values ​​of each node , and after the broadcast Normalization is performed and the data is concatenated along the feature dimensions to generate a shape of... The input tensor, where The time step of the input sequence. 3 represents the total number of dam seepage pressure monitoring nodes, and 3 represents the characteristic channel; The process for inverting the dominant seepage path inside the dam body includes: Obtain and calculate the mean of the attention weight matrix obtained during model training over the time dimension. , as the global correlation strength matrix; Calculate the absolute value of the rate of change of seepage pressure at each measuring point at the predicted time. As a node activity vector ; Define connecting edges Leakage risk index for: , For connecting edges The mean of the attention weight matrix over the time dimension. For nodes The activity vector; Screening out leakage risk index Connection edges exceeding a preset threshold serve as the dominant leakage channels, and paths are constructed in three-dimensional space.

2. The method for reconstructing the seepage field and inverting seepage channels in a dam as described in claim 1, characterized in that, The spatiotemporal graph attention network model includes a physically constrained spatial layer and a temporal evolution layer, wherein: A physical constraint space layer is used to extract spatial features according to the calculation formula; The time evolution layer employs a gated recurrent unit network to take the feature sequence output from the physical constraint space layer as input and output the predicted seepage pressure value for the next time step.

3. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 2.

Images