A dike emergency anti-seepage reinforcement treatment method combined with water flow force numerical simulation

Abstract

This invention discloses a method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation. The method collects geological and real-time hydrological data of the dikes to construct a standardized dataset; a first model based on the multi-field coupling theory of flow-seepage-stress outputs risk indicators; a second model optimized using a feature bag strategy with random forests transforms the risk indicators into risk levels; a third model constructed using a dual-channel convolutional neural network matches and optimizes seepage prevention and reinforcement schemes, outputting the optimal scheme and construction parameters; the reinforcement effect is monitored in real time, and the weights of the indicators are determined and a comprehensive score of the reinforcement effect is calculated through a reinforcement effect evaluation system, which is then updated to a historical case dataset. This method, through the synergistic operation of the three models, obtains accurately matched seepage prevention and reinforcement schemes, improving the accuracy and timeliness of emergency seepage prevention and reinforcement of dikes. Furthermore, by updating the database through the reinforcement effect evaluation system, the long-term effectiveness of the schemes is ensured, making it widely applicable.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering technology, specifically to a method for emergency seepage prevention and reinforcement of dikes that combines hydrodynamic numerical simulation with the application of machine learning models. Background Technology

[0002] With the deep integration and application of the Internet of Things, computer vision, and intelligent algorithms in the field of water conservancy engineering, emergency seepage prevention and reinforcement technology for dikes is upgrading towards precision and intelligence. The high-frequency flood disasters caused by extreme weather further impose stringent requirements on the stability of dike seepage prevention and the efficiency of emergency response. Dikes are critical infrastructure for resisting flood disasters, and under flood conditions, excessive seepage can easily lead to piping, deep cracks, and breaches. Rapid and precise emergency seepage prevention and reinforcement technology has become a core support for reducing disaster losses. Currently, several technical solutions focusing on intelligent monitoring and intelligent assessment have been developed in related fields; however, there is still a gap in the application of machine learning algorithms to efficiently and accurately generate emergency seepage prevention and reinforcement solutions for dikes.

[0003] Chinese patent (publication number CN119919418A) discloses a method for detecting surface defects of dikes based on deep learning image processing, which achieves efficient and accurate identification of surface defects of dikes through deep learning networks.

[0004] Chinese patent (publication number CN118709033B) discloses a method for recommending a defect detection and diagnosis scheme for a cutoff wall based on machine learning, which uses PSO and SVM algorithms to diagnose defects in the cutoff wall.

[0005] Chinese patent (publication number CN118709033B) discloses a riverbank reinforcement device and method in water conservancy construction, which realizes reinforcement scheme optimization and automated construction through sensor network and cloud interconnection.

[0006] As can be seen from the above typical technical solutions, the current exploration of artificial intelligence in water conservancy projects is limited to detection and evaluation. The generation of reinforcement and treatment solutions for dikes has not yet broken through the technical bottleneck of generating complete reinforcement solutions using machine learning algorithms. As a result, when dealing with the needs of emergency seepage prevention and reinforcement of dikes under complex hydrogeological conditions, the generation and implementation of reinforcement solutions rely on human experience and sensor networks to complete the solution design, which has problems such as high cost, slow response, and strong subjectivity.

[0007] To address the aforementioned shortcomings, there is an urgent need in the field to develop an emergency seepage prevention and reinforcement method that integrates multi-field coupled numerical simulation, intelligent risk classification, precise scheme matching, and quantitative effect evaluation. Summary of the Invention

[0008] Based on the above-mentioned technical problems, this application discloses an emergency seepage prevention and reinforcement method for dikes that combines hydrodynamic numerical simulation, specifically including:

[0009] Collect levee data, and construct a standardized dataset after preprocessing. The levee data includes levee geological data and real-time hydrological data.

[0010] Based on a standardized dataset, the first model outputs the risk index of the dike. The first model is constructed based on the hydrodynamic equation, Darcy's law and the effective stress principle, and the key parameters of the coupled model are inverted and optimized through a BP neural network.

[0011] Based on the risk indicators of the dikes, the second model is used to output the risk level of the dikes. The second model is constructed based on the random forest algorithm optimized by the feature bag strategy.

[0012] Based on the risk level results, an initial scheme is matched from a pre-set anti-seepage reinforcement scheme library using a third model, and the optimal anti-seepage reinforcement scheme is calculated. The third model is constructed based on a dual-channel convolutional neural network.

[0013] A reinforcement effect evaluation index system was constructed, and a comprehensive score of the reinforcement effect was output by combining real-time monitoring data and uploaded to the historical case dataset.

[0014] Preferably, the first model is a multi-field coupled model of water flow-seepage-stress, which is constructed based on three theoretical equations to realize the coupling and linkage of water flow dynamics, seepage, and stress-strain.

[0015] Boundary conditions are determined and parameters are initialized using real-time hydrological data.

[0016] A multi-field coupled numerical solution module is constructed, and multi-field coupled calculation is realized through numerical algorithms combined with software linkage;

[0017] Construct a BP neural network parameter optimization module to inversely optimize the key parameters of the coupled model;

[0018] The first model is formed by integrating a multi-field coupled numerical solution module and a BP neural network parameter optimization module to achieve dynamic prediction of risk indicators.

[0019] Preferably, the multi-field coupled numerical solution module specifically comprises:

[0020] The numerical algorithm of the multi-field coupled numerical solution module includes the hydrodynamic control equations, seepage control equations, and stress-strain control equations. The hydrodynamic control equations employ the Reynolds-averaged Navier-Stokes equations, combined with... The turbulence model describes the velocity and water level distribution of the water flow around the dike, and the formula is:

[0021]

[0022] in, , For the velocity component, For pressure, For dynamic viscosity, For the gravitational component, For source terms;

[0023] The seepage control equation, based on Darcy's law and the continuity equation, describes the movement of pore water within the dike. The formula is as follows:

[0024]

[0025] in, Let be the permeability tensor. For the water head, This refers to the volumetric water content.

[0026] The stress-strain governing equations, based on the effective stress principle and the Mohr-Coulomb constitutive model, describe the deformation and stability of the levee soil. The formula is as follows:

[0027]

[0028] in, For effective stress, For the total stress, Pore ​​water pressure, The Kronecker function;

[0029] The software linkage of the multi-field coupled numerical solution module uses FLOW-3D software. The finite volume method (FVM) is used to discretely determine the control equations to solve for the water flow velocity; the finite element method (FEM) is used to discretely determine the seepage gradient and seepage flow rate; and the explicit finite difference method is used to solve for the anti-sliding stability safety factor.

[0030] Preferably, the integrated multi-field coupled numerical solution module and the BP neural network parameter optimization module specifically comprises:

[0031] The real-time collected 3D geological and hydrological data of the dikes are input into the trained BP neural network, which outputs optimized key parameters.

[0032] The optimized parameters are substituted into the multi-field coupled numerical solution module to perform coupled iterative calculations, outputting accurate risk indicators, including seepage gradient. seepage flow Anti-slip stability safety factor Water flow velocity Based on the above data, the risk index is calculated using the following formula:

[0033]

[0034] in, The risk index for seepage in dikes. To allow for seepage gradient in the embankment soil, To allow for seepage flow, To allow for a safety factor for anti-skid stability, The critical velocity for piping, This refers to the timeliness coefficient;

[0035] Real-time monitoring data is fed back to the BP neural network, the parameters are retrained and optimized, the calculation results of the coupled model are updated, and dynamic prediction of risk indicators is achieved.

[0036] Preferably, the second model is as follows:

[0037] The core risk indicators output by the first model and the intermediate parameters obtained from the calculation were selected for dataset preparation and preprocessing.

[0038] Based on historical data on levee emergency response, a training set was constructed, and a random forest model was trained.

[0039] The preprocessed dataset is input into the trained random forest model, which outputs the risk level of each area of ​​the dike.

[0040] Preferably, the training of the random forest model specifically involves: inputting the training set into the random forest model, training decision trees one by one, and generating risk level prediction results independently for each decision tree;

[0041] The optimal parameter combination is selected by traversing the combination of parameters such as the number of decision trees, maximum depth, and number of feature samples using a grid search method, with the prediction accuracy of the test set as the optimization objective.

[0042] A majority voting mechanism is used to merge the prediction results of all decision trees. Specifically, the risk level labels output by each decision tree are statistically analyzed, and the label that appears most frequently is taken as the predicted risk level of the dike area. The formula is as follows:

[0043]

[0044] in, The risk level label predicted by the model, To find the independent variable corresponding to the maximum value, This is an indicator function; it returns 1 if the condition within the parentheses is true, and 0 otherwise. For the first A decision tree, Risk level;

[0045] After training, the model performance is verified using a test set. The core evaluation metrics are accuracy, precision, and recall.

[0046] Preferably, the third model is specifically:

[0047] The input feature set of the third model includes the hazard features output by the first model, the risk level output by the second model, and the levee foundation features; the output adopts a dual-branch structure, which outputs the category of the optimal scheme in the preset scheme library and the key construction parameters of the corresponding scheme respectively.

[0048] The third model is built on a convolutional neural network (CNN). The structure design adopts the classic architecture of convolutional layers, pooling layers, fully connected layers, and dual output layers. The dual outputs are the scheme type branch and the parameter optimization branch, respectively.

[0049] The third model matches an initial scheme from a pre-set anti-seepage reinforcement scheme library based on input features, and mines the deep correlation between features and scheme parameters through network training, so as to achieve accurate matching of scheme types and adaptive optimization of key construction parameters.

[0050] Preferably, the precise matching of the implementation scheme type and the adaptive optimization of key construction parameters specifically involve:

[0051] After receiving the risk level of the second model, the hazard parameters of the first model, and the feature set constructed from the basic data of the dike, the matching probability of each preset scheme is output through the scheme type branch, and the scheme with the highest probability is selected as the initial reinforcement scheme.

[0052] The normalized values ​​of the key parameters corresponding to the initial solution are output through the parameter optimization branch. The formula is as follows:

[0053]

[0054] in, For the first Normalized values ​​of key parameters, The first output of the CNN model Predicted values ​​of key parameters, For parameter normalization output, , The first The historical maximum and minimum boundary values ​​of each parameter For the first Historical measured values ​​of each construction parameter This is a precision verification indicator function used to filter parameters that meet the precision requirements. It outputs 1 if the requirements are met, and 0 if they are not. This is a preset error threshold;

[0055] The actual construction parameters are obtained through inverse normalization, and the optimal seepage prevention and reinforcement scheme is generated.

[0056] Receive real-time monitoring data during the construction process, supplement the training dataset, retrain the model regularly, and continuously optimize the accuracy of scheme matching and parameter prediction.

[0057] Preferably, the reinforcement effect evaluation index system is as follows: the reinforcement effect evaluation index system determines the weight of each index through the analytic hierarchy process (AHP), and obtains the comprehensive evaluation value through weighted calculation;

[0058] Regularly integrate real-time monitoring data into the database, update the reinforcement effect evaluation results and seepage prevention capacity boundary parameters, and synchronously apply the updated historical dataset to the training and optimization of the first, second and third models to improve the model prediction accuracy.

[0059] Preferably, the reinforcement effect evaluation index is divided into core evaluation index and auxiliary evaluation index. The core evaluation index is used to directly characterize the seepage prevention and reinforcement effect, including seepage slope control rate, seepage flow reduction rate, effective stress increase rate and anti-sliding stability safety factor compliance rate.

[0060] Auxiliary evaluation indicators are used to characterize the durability of reinforcement effect and construction quality, including the integrity index of the impermeable body, long-term deformation rate and seepage stability duration;

[0061] The weights of the core evaluation indicators were determined using the Analytic Hierarchy Process (AHP), while the weights of the auxiliary evaluation indicators were preset fixed values.

[0062] Compared with the prior art, the technical solution of this application has the following technical effects:

[0063] This invention constructs a technical system for hazard prediction, solution generation, parameter optimization, effect evaluation, and data iteration through the synergistic linkage of the first, second, and third models, breaking through the limitations of existing technologies that only focus on intelligent monitoring and defect diagnosis.

[0064] This invention directly outputs an adapted emergency seepage prevention and reinforcement scheme and optimal construction parameters through a third model of a dual-channel convolutional neural network, without the need for human experience intervention. This significantly improves the speed of emergency response and the scientific nature of decision-making, and effectively solves the problems of delayed generation and strong subjectivity in existing technical solutions.

[0065] This invention constructs a dynamic mapping relationship between risk characteristics, scheme types, and construction parameters through a third model, thereby achieving precise matching between reinforcement schemes and the degree of danger and geological and hydrological conditions, ensuring the engineering adaptability of construction parameters, and solving the defect of poor adaptability of existing technical schemes.

[0066] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0067] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0069] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0070] Figure 1 This is a flowchart illustrating an emergency seepage prevention and reinforcement method for dikes that incorporates hydrodynamic numerical simulation.

[0071] Figure 2 This is a general architecture diagram of an emergency seepage prevention and reinforcement method for dikes that combines hydrodynamic numerical simulation.

[0072] Figure 3 This is a diagram of the neural network structure of the first model;

[0073] Figure 4 This is a diagram of the random forest architecture for the second model;

[0074] Figure 5 This is a diagram of the neural network structure of the third model;

[0075] Figure 6 Heat map of seepage risk for a Class II levee section that underwent emergency seepage prevention and reinforcement using this method;

[0076] Figure 7 The image shows a comparison of the reinforcement scheme before and after the application of this method.

[0077] Figure 8 A data comparison chart of various methods when a piping failure is triggered in a simulation device;

[0078] Figure 9This is a data comparison chart showing the application of various methods in simulated defense under fluctuating environments. Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0080] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0081] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0082] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0083] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0084] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0085] Example 1 mainly describes an emergency seepage prevention and reinforcement method for dikes based on hydrodynamic numerical simulation, such as... Figure 1 , Figure 2 As shown, it specifically includes:

[0086] Collect levee data, and construct a standardized dataset after preprocessing. The levee data includes levee geological data and real-time hydrological data.

[0087] Based on a standardized dataset, the first model outputs the risk index of the dike. The first model is constructed based on the hydrodynamic equation, Darcy's law and the effective stress principle, and the key parameters of the coupled model are inverted and optimized through a BP neural network.

[0088] Based on the risk indicators of the dikes, the second model is used to output the risk level of the dikes. The second model is constructed based on the random forest algorithm.

[0089] Based on the risk level results, an initial scheme is matched from a pre-set anti-seepage reinforcement scheme library, input into a third model, and the optimal anti-seepage reinforcement scheme is output. The third model is constructed based on a convolutional neural network.

[0090] A reinforcement effect evaluation index system was constructed. By combining real-time monitoring data and the prediction results of the first model, the seepage prevention capacity boundary of the dike under different flood recurrence periods was determined, and the historical case dataset was improved.

[0091] Furthermore, such as Figure 3 The first model neural network structure diagram shown is as follows: The first model is a multi-field coupling model of water flow-seepage-stress, which is constructed based on three theoretical equations to realize the coupling and linkage of water flow dynamics, seepage, stress and strain.

[0092] Boundary conditions are determined and parameters are initialized using real-time hydrological data.

[0093] A multi-field coupled numerical solution module is constructed, and multi-field coupled calculation is realized through numerical algorithms combined with software linkage;

[0094] Construct a BP neural network parameter optimization module to inversely optimize the key parameters of the coupled model;

[0095] The first model is formed by integrating a multi-field coupled numerical solution module and a BP neural network parameter optimization module to achieve dynamic prediction of risk indicators.

[0096] Furthermore, the boundary conditions and parameter initialization are as follows:

[0097] Input real-time hydrological data as the upstream inflow boundary and downstream outflow boundary; set the surface of the dike as the solid wall boundary.

[0098] The surface water head of the dike calculated by the flow model is used as the upper boundary of the seepage model; the bottom of the dike foundation is set as an impermeable boundary; and the lateral boundary is set as a constant head or constant flow boundary.

[0099] The self-weight of the embankment is the volume force boundary; the top and slope surfaces of the embankment are set as free boundaries; and the bottom of the embankment foundation is set as a fixed displacement boundary.

[0100] Furthermore, the multi-field coupled numerical solution module specifically includes:

[0101] The numerical algorithm of the multi-field coupled numerical solution module includes the hydrodynamic control equations, seepage control equations, and stress-strain control equations. The hydrodynamic control equations employ the Reynolds-averaged Navier-Stokes equations, combined with... The turbulence model describes the velocity and water level distribution of the water flow around the dike, and the formula is:

[0102]

[0103] in, , For the velocity component, For pressure, For dynamic viscosity, For the gravitational component, For source terms;

[0104] The seepage control equation, based on Darcy's law and the continuity equation, describes the movement of pore water within the dike. The formula is as follows:

[0105]

[0106] in, Let be the permeability tensor. For the water head, This refers to the volumetric water content.

[0107] The stress-strain governing equations, based on the effective stress principle and the Mohr-Coulomb constitutive model, describe the deformation and stability of the levee soil. The formula is as follows:

[0108]

[0109] in, For effective stress, For the total stress, Pore ​​water pressure, The Kronecker function;

[0110] The software linkage of the multi-field coupled numerical solution module uses FLOW-3D software. The finite volume method (FVM) is used to discretely determine the control equations to solve for the water flow velocity; the finite element method (FEM) is used to discretely determine the seepage gradient and seepage flow rate; and the explicit finite difference method is used to solve for the anti-sliding stability safety factor.

[0111] Furthermore, the coupled solution process is as follows:

[0112] Step 1: Calculate the initial flow field using the flow model and output the water head at each node on the embankment surface;

[0113] Step 2: Import the head boundary into the seepage model and calculate the initial seepage field;

[0114] Step 3: Import pore water pressure into the stress model and calculate the effective stress and soil deformation;

[0115] Step 4: Correct the permeability coefficient based on the soil deformation;

[0116] Step 5: Repeat steps 1-4 until the rate of change of seepage flow and deformation in two adjacent iterations is less than the preset threshold, and output the converged coupled calculation results.

[0117] Furthermore, the construction and training of the BP neural network parameter optimization module are as follows:

[0118] The core function of the BP neural network is to invert and optimize the key parameters of the coupled model, solving the problems of traditional parameter assignment relying on experience and large prediction errors. The number of input layer nodes in its neural network structure design is the feature dimension of the standardized dataset, including soil type, particle size distribution, porosity, real-time water level difference, and flow velocity.

[0119] The hidden layers are set to 3, with the number of nodes in each layer set to 1.5 times the number of nodes in the input layer. The activation function is the Sigmoid function, with the following formula: To achieve nonlinear mapping;

[0120] The number of nodes in the output layer is a key parameter dimension of the coupled model to be optimized;

[0121] The specific data set partitioning and training are as follows:

[0122] Historical case data from the standardized dataset were selected and divided into training and test sets in a 7:3 ratio.

[0123] The training set input feature parameters are fed into the BP neural network, and the prediction parameters of the output layer are calculated through forward propagation.

[0124] Using the measured parameters as labels, the error between the predicted and measured values ​​is calculated, and the mean squared error (MSE) is used as the loss function. The formula is as follows: ;

[0125] The weights and biases of each layer are adjusted in the direction of error reduction using the backpropagation algorithm.

[0126] Furthermore, the integration of the multi-field coupled numerical solution module and the BP neural network parameter optimization module specifically includes:

[0127] The real-time collected 3D geological and hydrological data of the dikes are input into the trained BP neural network, which outputs optimized key parameters.

[0128] The optimized parameters are substituted into the multi-field coupled numerical solution module to perform coupled iterative calculations, outputting accurate risk indicators, including seepage gradient. seepage flow Anti-slip stability safety factor Water flow velocity Based on the above data, the risk index is calculated using the following formula:

[0129]

[0130] in, The risk index for seepage in dikes. To allow for seepage gradient in the embankment soil, To allow for seepage flow, To allow for a safety factor for anti-skid stability, The critical velocity for piping, This refers to the timeliness coefficient;

[0131] Real-time monitoring data is fed back to the BP neural network, the parameters are retrained and optimized, the calculation results of the coupled model are updated, and dynamic prediction of risk indicators is achieved.

[0132] Furthermore, such as Figure 4 The architecture diagram of the second model is shown below. The second model is specifically as follows:

[0133] The core risk indicators output by the first model and the intermediate parameters obtained from the calculation were selected for dataset preparation and preprocessing.

[0134] Based on historical data on levee emergency response, a training set was constructed, and a random forest model was trained.

[0135] The preprocessed dataset is input into the trained random forest model, which outputs the risk level of each area of ​​the dike.

[0136] Furthermore, the dataset preparation and preprocessing specifically involve integrating two parts of data: first, the simulation calculation data of the first model, which includes risk indicators and corresponding risk levels under different hydrological and geological conditions; and second, historical dike emergency response case data, which includes measured risk indicators, actual risk levels, and data verifying the response effectiveness.

[0137] The SMOTE algorithm is used to address the class imbalance problem in the dataset. The dataset is divided into training and test sets in a 7:3 ratio. The input features are standardized by min-max normalization to avoid the impact of differences in scale on the model training effect.

[0138] Furthermore, the formula for configuring the structure parameters of the random forest model is as follows:

[0139]

[0140] in, It is a random forest model. For the first A decision tree, The number of decision trees;

[0141] The core parameters of the model are configured as follows: the number of decision trees is set to 100, determined using a grid search method; the maximum depth of each decision tree is set to 10 to prevent overfitting; a feature bagging strategy is adopted, where 3 features are randomly selected from 5 input features during the training of each decision tree to improve the model's generalization ability; and the Gini coefficient is used as the basis for evaluating feature importance and node splitting. The formula for calculating the Gini coefficient is as follows:

[0142]

[0143] in, For the sample to belong to the first The smaller the Gini coefficient, the higher the sample purity and the better the splitting effect.

[0144] The samples were obtained using bootstrap sampling, with 70% of the samples randomly selected from the training set for training a single decision tree, and the remaining 30% of the samples used as out-of-bag data for internal model validation.

[0145] Furthermore, the process of training the random forest model is as follows: the training set is input into the random forest model, the decision trees are trained one by one, and each decision tree independently generates the risk level prediction result;

[0146] The optimal parameter combination is selected by traversing the combination of parameters such as the number of decision trees, maximum depth, and number of feature samples using a grid search method, with the prediction accuracy of the test set as the optimization objective.

[0147] A majority voting mechanism is used to merge the prediction results of all decision trees. Specifically, the risk level labels output by each decision tree are statistically analyzed, and the label that appears most frequently is taken as the predicted risk level of the dike area. The formula is as follows:

[0148]

[0149] in, The risk level label predicted by the model, To find the independent variable corresponding to the maximum value, This is an indicator function; it returns 1 if the condition within the parentheses is true, and 0 otherwise. For the first A decision tree, Risk level;

[0150] After training, the model performance is verified using a test set. The core evaluation metrics are accuracy, precision, and recall.

[0151] Furthermore, such as Figure 5 The neural network structure diagram of the third model is shown below. The third model is specifically as follows:

[0152] The input feature set of the third model includes the hazard features output by the first model, the risk level output by the second model, and the levee foundation features; the output adopts a dual-branch structure, which outputs the category of the optimal scheme in the preset scheme library and the key construction parameters of the corresponding scheme respectively.

[0153] The third model is built on a convolutional neural network (CNN). The structure design adopts the classic architecture of convolutional layers, pooling layers, fully connected layers, and dual output layers. The dual outputs are the scheme type branch and the parameter optimization branch, respectively.

[0154] The third model matches an initial scheme from a pre-set anti-seepage reinforcement scheme library based on input features, and mines the deep correlation between features and scheme parameters through network training, so as to achieve accurate matching of scheme types and adaptive optimization of key construction parameters.

[0155] Furthermore, the training and optimization process of the third model is as follows:

[0156] A multi-dimensional fusion feature vector is constructed, covering 3 core features and a total of 10 dimensions, specifically: the risk level output by the second model, which is numerically encoded; the seepage gradient, seepage flow, and pore water pressure output by the first model; and basic features, including the type of levee soil layer, permeability coefficient, cohesion, internal friction angle, water level difference, and water flow velocity, which constitute the input vector. ;

[0157] The core structural parameters of the third model are configured as follows:

[0158] The input layer has 10 nodes, corresponding to 10-dimensional input features. It reconstructs the one-dimensional feature vector into a 2×5 two-dimensional feature map, which is suitable for the feature extraction requirements of convolution operations.

[0159] The convolutional layer consists of two layers. The first layer has a kernel size of 3×3 and a number of 32 kernels, and uses ReLU as the activation function to achieve basic feature mining. The second layer has a kernel size of 2×2 and a number of 64 kernels to deepen the high-dimensional feature representation.

[0160] Two max pooling layers are set for the convolutional layers, with a pooling kernel size of 2×2 and a stride of 2, to achieve feature dimensionality reduction and reduce the number of parameters and the risk of overfitting.

[0161] Two fully connected layers are set up, with 128 nodes in the first layer and 64 nodes in the second layer, to achieve non-linear mapping of high-dimensional features;

[0162] The output layer has a dual-branch structure, where the number of nodes in the scheme type branch is equal to the number of schemes in the preset scheme library, and the activation function is Softmax; the number of nodes in the parameter optimization branch is equal to the total number of key parameters for each scheme, and the activation function is Sigmoid.

[0163] The dataset of the third model is divided into training and validation sets in an 8:2 ratio, and the input features are processed using Z-Score standardization to eliminate differences in scale.

[0164] Using a combined loss function, the cross-entropy loss from combined scenario type prediction and the mean squared error loss from parameter prediction are combined, as shown in the formula:

[0165]

[0166] in, For the combined loss function, These are the weighting coefficients. For the first The predicted probability of each option, For the first The logarithm of the predicted probability of each option. For the first The actual measured values ​​of each construction parameter. For the first Model prediction values ​​for each construction parameter;

[0167] The training optimization process uses the Adam optimizer, with an initial learning rate of 0.001 and a learning rate decay strategy (decaying to 0.9 of the original learning rate every 100 iterations). The number of training iterations is set to 500. Training stops when the validation set loss function value does not decrease for 20 consecutive iterations, and the optimal model parameters are saved.

[0168] Furthermore, the process of achieving precise matching of scheme types and adaptive optimization of key construction parameters is as follows:

[0169] After receiving the risk level of the second model, the hazard parameters of the first model, and the feature set constructed from the basic data of the dike, the matching probability of each preset scheme is output through the scheme type branch, and the scheme with the highest probability is selected as the initial reinforcement scheme.

[0170] The normalized values ​​of the key parameters corresponding to the initial solution are output through the parameter optimization branch. The formula is as follows:

[0171] in, Let be the normalized value of the i-th key parameter. This represents the predicted value of the i-th key parameter output by the CNN model. For parameter normalization output, , These are the historical maximum and minimum boundary values ​​of the i-th parameter, respectively. For the first Historical measured values ​​of each construction parameter This is a precision verification indicator function used to filter parameters that meet the precision requirements. It outputs 1 if the requirements are met, and 0 if they are not. This is a preset error threshold;

[0172] The actual construction parameters are obtained through inverse normalization, and the optimal seepage prevention and reinforcement scheme is generated.

[0173] Receive real-time monitoring data during the construction process, supplement the training dataset, retrain the model regularly, and continuously optimize the accuracy of scheme matching and parameter prediction.

[0174] Furthermore, the reinforcement effect evaluation index system is as follows: the weight of each index is determined by the analytic hierarchy process (AHP), and the comprehensive evaluation value is obtained by weighted calculation.

[0175] Regularly integrate real-time monitoring data into the database, update the reinforcement effect evaluation results and seepage prevention capacity boundary parameters, and synchronously apply the updated historical dataset to the training and optimization of the first, second and third models to improve the model prediction accuracy.

[0176] Furthermore, the specific evaluation indicators for reinforcement effect are as follows: The evaluation indicators for reinforcement effect are divided into core evaluation indicators and auxiliary evaluation indicators. Among them, the core evaluation indicators are used to directly characterize the seepage prevention and reinforcement effect, including the seepage slope control rate, seepage flow reduction rate, effective stress increase rate and the compliance rate of anti-sliding stability safety factor.

[0177] Auxiliary evaluation indicators are used to characterize the durability of reinforcement effect and construction quality, including the integrity index of the impermeable body, long-term deformation rate and seepage stability duration;

[0178] The weights of the core evaluation indicators were determined using the Analytic Hierarchy Process (AHP), while the weights of the auxiliary evaluation indicators were preset fixed values.

[0179] This embodiment details a method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation. A standardized dataset is constructed by collecting geological and real-time hydrological data of the dikes. A first model based on the multi-field coupling theory of water flow, seepage, and stress outputs risk indicators. A second model using a random forest optimized by a feature bag strategy transforms the risk indicators into risk levels. A third model constructed using a dual-channel convolutional neural network matches and optimizes seepage prevention and reinforcement schemes, outputting the optimal scheme and construction parameters. The reinforcement effect is monitored in real time, and the weights of the indicators are determined and a comprehensive score for the reinforcement effect is calculated through a reinforcement effect evaluation system, which is then synchronously updated to the historical case dataset.

[0180] Example 2, based on Example 1, describes in detail an experiment using the method of the present invention to carry out emergency seepage prevention and reinforcement treatment in a Class 2 levee section, such as... Figure 6 The heat map showing the seepage risk of a Class 2 levee section is shown. This levee is 8.2m high and 6m wide at the crest. During the 2024 flood season, affected by heavy rainfall upstream, the measured water level in this section reached 16.3m, exceeding the warning level by 1.4m. The continuous high water level caused multiple piping incidents at the levee toe, with the largest piping point having a diameter of 30cm and an initial seepage flow of 0.04m³ / s. The situation was rapidly escalating. The implementation process for emergency seepage prevention and reinforcement is as follows:

[0181] Measurements of the foundation engineering conditions of the embankment section revealed that the soil layer of the embankment body is silty loam with a permeability coefficient of [missing information]. =1.8× cm / s, cohesion 17kPa, internal friction angle The foundation soil is sandy loam with a permeability coefficient of [missing information]. = cm / s, cohesion 8kPa, internal friction angle The land 50m outside the toe of the dike is farmland with no important structures, which is suitable for the operation of large construction machinery.

[0182] The parameters of the embankment body and foundation soil layers were obtained through portable dynamic probing and water injection tests; the current water level was measured to be 16.3m and the water flow velocity was 0.9m / s using a water level gauge and a flow velocity meter, and the water level difference Δh = 6.5m was calculated.

[0183] Fiber optic sensors, pore water pressure gauges, and seepage flow monitors were deployed to obtain the following data: pore water pressure at the dike toe u=62kPa, seepage flow at the piping point Q=0.04m³ / s, dike settlement rate 0.5mm / h, and horizontal displacement rate 0.3mm / h.

[0184] The historical case data was compiled by calling up 120 cases of emergencies from the flood season monitoring data of the dike section over the past 10 years and 20 cases of successful handling from the historical reinforcement construction records.

[0185] Set the boundary conditions for the multi-field coupled water flow-seepage-stress model, initialize the BP neural network of the first model, with 12 neurons in the input layer corresponding to 12 basic parameters; 3 hidden layers with 40, 20 and 16 neurons respectively; and 4 neurons in the output layer corresponding to 4 core risk indicators.

[0186] Using historical case data as training samples, the key parameters of the multi-field coupling model were inverted and optimized through a BP neural network. The model's prediction mean square error (MSE) was 0.003, which met the accuracy requirements.

[0187] Through multi-field coupled calculations and BP optimization, the core risk indicator is output: actual permeability gradient. Greater than the permeability gradient threshold Actual seepage flow greater than the seepage flow threshold Anti-slip stability safety factor =1.18, which meets the safety factor threshold. Flow velocity at piping point Less than the critical flow velocity .

[0188] The feature vector output by the first model is input into the trained random forest model. The model makes an ensemble voting decision and outputs a risk level of Level I emergency (label=2).

[0189] Substitute the above parameters into the risk index calculation formula:

[0190]

[0191] The calculated risk index was 7.391, which is greater than the preset threshold of 6.0. The situation was determined to be Level I, requiring immediate emergency reinforcement measures. This is consistent with the preliminary judgment result of the second model, verifying the reliability of the model in judging the urgency of the situation.

[0192] Based on the risk information output by the first two models, a 16-dimensional input feature vector is constructed, which includes risk level, 4 risk indicators, 7 geological parameters, and 4 hydrological parameters. This vector is then input into the third model to complete scheme matching and parameter optimization.

[0193] The solution matching branch of the third model outputs the matching probability of a preset solution library through the Softmax activation function:

[0194] High-pressure jet grouting, reverse filter layer, and decompression well have a matching probability of 0.93.

[0195] Double-row mixing pile seepage prevention wall and curtain grouting, matching probability 0.05;

[0196] The composite geomembrane was laid and the clay was compacted, with a matching probability of 0.02.

[0197] High-pressure jet grouting, reverse filter layer, and decompression well, which have the highest matching probability, were selected as the core reinforcement schemes.

[0198] The parameter optimization branch of the third model, after CNN forward propagation and Sigmoid normalization, outputs normalized parameters:

[0199] High-pressure jet grouting pressure: 0.65; grouting lifting speed: 0.4; pressure relief well spacing: 0.3.

[0200] By combining the inverse normalization formula and based on historical construction parameter ranges (grouting pressure 20-30 MPa, lifting speed 5-10 cm / min, pressure relief well spacing 5-8 m), the actual construction parameters are reconstructed:

[0201] The high-pressure jet grouting pressure is 0.65×(30-20)+20=26.5MPa, the grouting lifting speed is 0.4×(10-5)+5=7cm / min, and the pressure relief well spacing is 0.3×(8-5)+5=5.9m.

[0202] The optimized reinforcement scheme and construction parameters were imported into BIM software to generate a three-dimensional construction model. The high-pressure jet grouting holes were arranged in a quincunx pattern with a spacing of 1.5m and a depth of 8.5m, penetrating the sandy loam layer. The pressure relief wells were arranged with a spacing of 6m and a depth of 10m, with a filter pipe length of 3m. The boundary and compaction standard for the reverse filter layer were set at a compaction degree of ≥93%.

[0203] Based on the solution output by the third model, the construction team was organized to carry out emergency response, and the data in Table 1 below was obtained:

[0204] Table 1 Monitoring data after applying the reinforcement scheme

[0205] According to Table 1 and Figure 7 As shown in the comparison diagram of the reinforcement schemes, the scheme generated by this method meets the design requirements in terms of seepage flow, seepage gradient, anti-sliding stability safety factor, embankment settlement rate, seepage prevention integrity index, long-term deformation rate, and seepage stability duration. The seepage prevention stability of the embankment is significantly improved after treatment, which verifies the engineering applicability of this method.

[0206] The weights of the seven core evaluation indicators were determined using the analytic hierarchy process combined with expert scoring. The values ​​are 0.2, 0.2, 0.15, 0.15, 0.1, 0.1, and 0.1, corresponding to seepage flow, seepage gradient, anti-sliding stability safety factor, levee settlement rate, impermeable body integrity index, long-term deformation rate, and seepage stability duration, respectively. A comprehensive evaluation value is calculated using a comprehensive scoring model.

[0207]

[0208] The overall evaluation score E = 92.95 points. According to the evaluation criteria, the emergency seepage prevention and reinforcement effect is deemed excellent, and the seepage prevention stability of the dike meets the requirements for flood control safety.

[0209] This embodiment details the application of the emergency seepage prevention and reinforcement method for dikes, which successfully addressed a Class I piping hazard in a Class 2 dike section. The entire process, from accurate hazard prediction and intelligent solution generation to parameter optimization and adaptation and quantitative effect evaluation, was completed. After the treatment, the seepage prevention stability of the dike was significantly improved, and the overall evaluation results were excellent, verifying the scientific validity and engineering applicability of this method.

[0210] Example 3, based on Example 1 or 2, describes in detail the construction of a dike piping failure simulation device in a laboratory environment consistent with the aforementioned engineering scenario. Three mainstream existing methods were selected as comparison items, and parallel tests were conducted. The specific process is as follows:

[0211] A 1:20 scale model of the dike was constructed, with a dike height of 41cm and a top width of 30cm. Silty loam and sandy loam were used to replicate the soil parameters of the original project. A water level control system, a seepage flow monitoring system, and a displacement sensing system were configured.

[0212] The comparative testing methods include Machine Learning-Based Defect Detection Method (ML-ADM), Deep Learning-Based Detection Method (DL-DM), and Edge-Cloud Interconnected Control Method (ECICM). Among them, ML-ADM and DL-DM are only for detection and cannot directly generate reinforcement measurements, requiring manual assistance to match the solution.

[0213] Five core indicators were selected for comparison, including risk prediction accuracy, solution generation time, reinforcement parameter optimization accuracy, post-reinforcement hazard control success rate, and overall disposal cost.

[0214] A piping failure was triggered in the simulation device, with seepage flow rates randomly ranging from 0.0021 m³ / s (corresponding to 0.042 m³ / s in the prototype). Four methods were simultaneously initiated to address the issue, with each method tested 20 times. The average value was taken as the result, as shown in Table 2 below.

[0215] Table 2 Comparison of data from various methods when triggering piping hazards in the simulation device.

[0216] According to Table 2 and Figure 8 The data comparison chart of various methods when triggering piping in the simulation device shows that ML-ADM and DL-DM are more focused on detection and prediction, with high accuracy in risk prediction. However, due to the lack of reinforcement scheme generation function, they suffer from a complete process failure, strong subjectivity of human intervention, and inability to provide a completely environment-matching scheme within a limited time, resulting in long generation time and low parameter optimization accuracy. ECICM suffers from low parameter optimization accuracy and low hazard control level because the sensor network has difficulty accurately locating risks on the simulated micro-dam. Our method, due to its full process adaptability, has significant improvements over existing methods in all aspects.

[0217] Random piping incidents were triggered in the simulation device, with each seepage flow rate randomly ranging from 0.0010 m³ / s to 0.0042 m³ / s, corresponding to the prototype flow rates of 0.021 m³ / s to 0.084 m³ / s. Each method was tested 20 times, and the data are shown in Table 3 below.

[0218] Table 3

[0219] According to Table 3 and Figure 9 As shown in the data comparison chart of the various methods used on the simulated dike under the fluctuating environment, it can be seen that the method of this method, under the scenario of dynamic fluctuation of seepage flow, still achieves an average risk prediction accuracy of 93.5% through the dynamic adjustment of reinforcement parameter weights by the dual-channel CNN module, with a fluctuation range of only ±1.2%, which is significantly better than the other methods that show a significant decline in performance.

[0220] This embodiment describes the laboratory simulation test of the performance of this method and the comparison with three other methods: ML-ADM, DL-DM, and ECICM. The test fully verifies that this method has significant advantages in prediction accuracy, treatment efficiency, optimization effect, and cost control, proving the scientific nature, efficiency, and engineering applicability of this method. It can be widely applied to the emergency treatment of various seepage hazards in dikes in plain areas.

[0221] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation, characterized in that, include: Collect levee data, and construct a standardized dataset after preprocessing. The levee data includes levee geological data and real-time hydrological data. Based on a standardized dataset, the first model outputs the risk index of the dike. The first model is constructed based on the hydrodynamic equation, Darcy's law and the effective stress principle, and the key parameters of the coupled model are inverted and optimized through a BP neural network. Based on the risk indicators of the dikes, the second model is used to output the risk level of the dikes. The second model is constructed based on the random forest algorithm optimized by the feature bag strategy. Based on the risk level results, an initial scheme is matched from a pre-set anti-seepage reinforcement scheme library using a third model, and the optimal anti-seepage reinforcement scheme is calculated. The third model is constructed based on a dual-channel convolutional neural network. A reinforcement effect evaluation index system was constructed, and a comprehensive score of the reinforcement effect was output by combining real-time monitoring data and uploaded to the historical case dataset.

2. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 1, characterized in that, The first model is specifically: the first model is a multi-field coupling model of water flow-seepage-stress, which is constructed based on three theoretical equations to realize the coupling and linkage of water flow dynamics, seepage, stress and strain; Boundary conditions are determined and parameters are initialized using real-time hydrological data. A multi-field coupled numerical solution module is constructed, and multi-field coupled calculation is realized through numerical algorithms combined with software linkage; Construct a BP neural network parameter optimization module to inversely optimize the key parameters of the coupled model; The first model is formed by integrating a multi-field coupled numerical solution module and a BP neural network parameter optimization module to achieve dynamic prediction of risk indicators.

3. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 2, characterized in that, The multi-field coupled numerical solution module is specifically as follows: The numerical algorithm of the multi-field coupled numerical solution module includes the hydrodynamic control equations, seepage control equations, and stress-strain control equations. The hydrodynamic control equations employ the Reynolds-averaged Navier-Stokes equations, combined with... The turbulence model describes the velocity and water level distribution of the water flow around the dike, and the formula is: in, , For the velocity component, For pressure, For dynamic viscosity, For the gravitational component, For source terms; The seepage control equation, based on Darcy's law and the continuity equation, describes the movement of pore water within the dike. The formula is as follows: in, Let be the permeability tensor. For the water head, This refers to the volumetric water content. The stress-strain governing equations, based on the effective stress principle and the Mohr-Coulomb constitutive model, describe the deformation and stability of the levee soil. The formula is as follows: in, For effective stress, For the total stress, Pore ​​water pressure, The Kronecker function; The software linkage of the multi-field coupled numerical solution module uses FLOW-3D software. The finite volume method (FVM) is used to discretely determine the control equations to solve for the water flow velocity; the finite element method (FEM) is used to discretely determine the seepage gradient and seepage flow rate; and the explicit finite difference method is used to solve for the anti-sliding stability safety factor.

4. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 2, characterized in that, The integrated multi-field coupled numerical solution module and BP neural network parameter optimization module are specifically as follows: The real-time collected 3D geological and hydrological data of the dikes are input into the trained BP neural network, which outputs optimized key parameters. The optimized parameters are substituted into the multi-field coupled numerical solution module to perform coupled iterative calculations, outputting accurate risk indicators, including seepage gradient. seepage flow Anti-slip stability safety factor Water flow velocity Based on the above data, the risk index is calculated using the following formula: in, The risk index for seepage in dikes. To allow for seepage gradient in the embankment soil, To allow for seepage flow, To allow for a safety factor for anti-skid stability, The critical velocity for piping, This refers to the timeliness coefficient; Real-time monitoring data is fed back to the BP neural network, the parameters are retrained and optimized, the calculation results of the coupled model are updated, and dynamic prediction of risk indicators is achieved.

5. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 1, characterized in that, The second model is as follows: The core risk indicators output by the first model and the intermediate parameters obtained from the calculation were selected for dataset preparation and preprocessing. Based on historical data on levee emergency response, a training set was constructed, and a random forest model was trained. The preprocessed dataset is input into the trained random forest model, which outputs the risk level of each area of ​​the dike.

6. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 5, characterized in that, The training of the random forest model specifically involves: inputting the training set into the random forest model, training decision trees one by one, and generating risk level prediction results independently for each decision tree; The optimal parameter combination is selected by traversing the combination of parameters such as the number of decision trees, maximum depth, and number of feature samples using a grid search method, with the prediction accuracy of the test set as the optimization objective. A majority voting mechanism is used to merge the prediction results of all decision trees. Specifically, the risk level labels output by each decision tree are statistically analyzed, and the label that appears most frequently is taken as the predicted risk level of the dike area. The formula is as follows: in, The risk level label predicted by the model, To find the independent variable corresponding to the maximum value, This is an indicator function; it returns 1 if the condition within the parentheses is true, and 0 otherwise. For the first A decision tree, Risk level; After training, the model performance is verified using a test set. The core evaluation metrics are accuracy, precision, and recall.

7. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation as described in claim 1, characterized in that, The third model is specifically as follows: The input feature set of the third model includes the hazard features output by the first model, the risk level output by the second model, and the levee foundation features; the output adopts a dual-branch structure, which outputs the category of the optimal scheme in the preset scheme library and the key construction parameters of the corresponding scheme respectively. The third model is built on a convolutional neural network (CNN). The structure design adopts the classic architecture of convolutional layers, pooling layers, fully connected layers, and dual output layers. The dual outputs are the scheme type branch and the parameter optimization branch, respectively. The third model matches an initial scheme from a pre-set anti-seepage reinforcement scheme library based on input features, and mines the deep correlation between features and scheme parameters through network training, so as to achieve accurate matching of scheme types and adaptive optimization of key construction parameters.

8. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 7, characterized in that, The precise matching of the implementation scheme type and the adaptive optimization of key construction parameters are specifically as follows: After receiving the risk level of the second model, the hazard parameters of the first model, and the feature set constructed from the basic data of the dike, the matching probability of each preset scheme is output through the scheme type branch, and the scheme with the highest probability is selected as the initial reinforcement scheme. The normalized values ​​of the key parameters corresponding to the initial solution are output through the parameter optimization branch. The formula is as follows: in, For the first Normalized values ​​of key parameters, The first output of the CNN model Predicted values ​​of key parameters, For parameter normalization output, , The first The historical maximum and minimum boundary values ​​of each parameter For the first Historical measured values ​​of each construction parameter This is a precision verification indicator function used to filter parameters that meet the precision requirements. It outputs 1 if the requirements are met, and 0 if they are not. This is a preset error threshold; The actual construction parameters are obtained through inverse normalization, and the optimal seepage prevention and reinforcement scheme is generated. Receive real-time monitoring data during the construction process, supplement the training dataset, retrain the model regularly, and continuously optimize the accuracy of scheme matching and parameter prediction.

9. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 1, characterized in that, The reinforcement effect evaluation index system is specifically as follows: the weight of each index is determined by the analytic hierarchy process (AHP), and the comprehensive evaluation value is obtained by weighted calculation. Regularly integrate real-time monitoring data into the database, update the reinforcement effect evaluation results and seepage prevention capacity boundary parameters, and synchronously apply the updated historical dataset to the training and optimization of the first, second and third models to improve the model prediction accuracy.

10. The method for emergency seepage prevention and reinforcement of dikes based on hydrodynamic numerical simulation according to claim 9, characterized in that, The reinforcement effect evaluation indicators are as follows: The reinforcement effect evaluation indicators are divided into core evaluation indicators and auxiliary evaluation indicators. The core evaluation indicators are used to directly characterize the seepage prevention and reinforcement effect, including seepage slope control rate, seepage flow reduction rate, effective stress increase rate and anti-sliding stability safety factor compliance rate. Auxiliary evaluation indicators are used to characterize the durability of reinforcement effect and construction quality, including the integrity index of the impermeable body, long-term deformation rate and seepage stability duration; The weights of the core evaluation indicators were determined using the Analytic Hierarchy Process (AHP), while the weights of the auxiliary evaluation indicators were preset fixed values.

Images