Earth and rockfill dam burst flow prediction method based on data enhancement and adaptive network

Abstract

The invention belongs to the technical field of machine learning, and discloses an earth and rockfill dam burst flow prediction method based on data enhancement and an adaptive network. Comprising the following steps that historical earth and rockfill dam outburst flow data are collected and preprocessed, and then historical earth and rockfill dam outburst flow data are obtained; designing and training a generative adversarial network, and expanding the historical earth and rockfill dam outburst flow data through the generative adversarial network to obtain an earth and rockfill dam outburst flow data set; performing feature extraction and weighting on the data in the data set to obtain outburst flow feature data; constructing a support vector machine model, and training the support vector machine model by using an adaptive particle swarm optimization algorithm to obtain a trained support vector machine model; according to the method, the training set is enhanced by using the generative adversarial network model and the attention mechanism, and the enhanced training set is input into the adaptive particle swarm optimization support vector machine model, so that accurate prediction of the earth and rockfill dam burst peak flow can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and more specifically, to a method for predicting earth-rock dam burst flow based on data enhancement and adaptive networks. Background Art

[0002] Accurate prediction of the peak flow of earth-rock dam burst is of great significance for flood prevention and disaster reduction, decision support and post-disaster recovery. In recent years, many scholars have tried to use various methods to predict the peak flow of earth-rock dam burst. Common methods include traditional hydrodynamic models, machine learning models and deep learning models;

[0003] However, the existing earth-rock dam burst peak flow prediction model has the following problems:

[0004] In the prediction of earth-rock dam burst flow, data scarcity is a common problem, especially for some rare or special burst events, historical data is often limited; traditional prediction methods are often unable to fully extract deep features in the data when faced with complex input data, resulting in limited prediction accuracy; traditional machine learning algorithms and models are prone to overfitting or low prediction accuracy when faced with high-dimensional and diverse input data.

[0005] In view of this, the present invention proposes a method for predicting earth-rock dam burst flow based on data enhancement and adaptive network to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for predicting earth-rock dam burst flow based on data enhancement and adaptive network, comprising the following steps:

[0007] S1. Collect and pre-process historical earth-rock dam burst flow data, and then obtain historical earth-rock dam burst flow data;

[0008] S2. Design and train a generative adversarial network, and expand the historical earth-rock dam burst flow data through the generative adversarial network to obtain an earth-rock dam burst flow data set; the specific method is: design a generative adversarial network, and generate earth-rock dam burst flow data similar to the historical earth-rock dam burst flow data in the training set by training the generator, wherein the generator generates earth-rock dam burst flow data through random noise; the discriminator determines whether the generated earth-rock dam burst flow data is real data, and compares the generated earth-rock dam burst flow data with the historical earth-rock dam burst flow data to obtain the true and false probability values; the generator continuously adjusts the generated earth-rock dam burst flow data through the true and false probability values ​​to make it consistent with the historical earth-rock dam burst flow data; the objective function of GAN is:

[0009] minG max D E x~pdata(x) [logD(x)] +E z~pz(z) [log(1-D(G(z)))];

[0010] In the formula, G(z) represents the generator, z represents random noise, x=G(z) represents the generated earth-rock dam burst flow data, D(x) represents the discriminator, which is used to output the probability that x belongs to the real data, pdata(x) represents the real data distribution, and pz(z) is the noise distribution of the generator input; E x~pdata(x) [logD(x)] represents the expectation of the discriminator’s judgment result on the historical earth-rock dam burst flow data x; E z~pz(z) [log(1-D(G(z)))] represents the expectation of the earth-rock dam burst flow data G(z) generated by the generator; min G The goal of the generator G is to minimize the discriminator's ability to identify; max D It means that the goal of the discriminator D is to maximize its ability to discriminate real data. Through iterative training, the generated earth-rock dam burst flow data is added to the historical earth-rock dam burst flow data to obtain the earth-rock dam burst flow data set.

[0011] S3. Use the Attention mechanism to extract features from the data in the earth-rock dam burst flow dataset, and weight the extracted features to obtain the characteristic data of the earth-rock dam burst flow, and divide the earth-rock dam burst flow characteristic data into a training set and a test set; the specific method is: the preset input query matrix Q, key matrix K, and value matrix V are Q∈R nq ×dk , K∈R nk×dk , V∈R nk×dv , where nq represents the number of queries, nk represents the number of keys, dk is the dimension of the query and key, dv is the dimension of the value, and R represents the prefix of the matrix dimension;

[0012] Calculate the attention score matrix Attention Scores by matrix multiplication, that is, Attention Scores=QK T , where K T is the transpose of the key matrix, and the attention score matrix has the size of nq×nk, which represents the similarity between each query and each key;

[0013] Scale the attention scores to obtain the scaled attention score matrix Scaled Scores, that is, Scaled Scores = QK T / √dk, where √dk is the square root of the bond dimension;

[0014] Apply the softmax function to each row in the scaled attention score matrix to obtain the attention weights Attention Weights, that is, Attention Weights=softmax(QK T / √dk);

[0015] Use the attention weight matrix and the value matrix to perform matrix multiplication to obtain the weighted summed output matrix Attention Output. The output matrix is ​​the characteristic data of earth-rock dam burst flow, that is, Attention Output = Attention Weights V;

[0016] The earth-rock dam burst flow characteristic data are divided into training set and test set, with 80% of the earth-rock dam burst flow characteristic data as the training set and 20% of the earth-rock dam burst flow characteristic data as the test set;

[0017] S4, constructing a support vector machine model, and using an adaptive particle swarm optimization algorithm to train the support vector machine model, thereby obtaining a trained support vector machine model; the specific method is: using an RBF kernel to construct a support vector machine model; obtaining the optimal parameters of the support vector machine model through an adaptive particle swarm optimization algorithm, the optimal parameters being the optimal solution of the regularization parameter C and the kernel function parameter γ; using the optimal parameters to train the support vector machine model until a preset maximum number of evolutions or an error threshold is reached, thereby obtaining a trained support vector machine model;

[0018] The method of obtaining the optimal parameters of the support vector machine model by using the adaptive particle swarm optimization algorithm is:

[0019] Initialize the particle swarm: set the size of the particle swarm, initialize the position and velocity of each particle, and the position of the particle represents a set of parameters, namely the regularization parameter C and the kernel function parameter γ;

[0020] Evaluate the fitness value of particles: define a fitness function to evaluate each particle and calculate the fitness value of each particle;

[0021] Update particle position and speed: Update the individual optimal position of each particle according to the fitness value. If the fitness value of the current particle is better than the fitness value of the global optimal position, update the global optimal position, introduce an adaptive mechanism to adjust the particle's speed and position update strategy, and use the updated position formula to calculate the new position and speed of each particle;

[0022] Repeatedly evaluate the fitness value of the particle and update the particle position until the predetermined number of iterations is reached or other stopping conditions are met;

[0023] S5. Perform performance evaluation and verification on the trained support vector machine model until it meets the preset performance evaluation index threshold;

[0024] S6. Collect the current earth-rock dam burst flow characteristic data, input the current earth-rock dam burst flow characteristic data into the trained support vector machine model, and obtain the current predicted burst peak flow.

[0025] Preferably, the method of collecting and preprocessing the data of historical earth-rock dam burst flow rates to obtain the data of historical earth-rock dam burst flow rates is:

[0026] Collect historical data on earth-rock dam breach discharges, including dam height, reservoir capacity, porosity, and corresponding earth-rock dam breach peak discharges;

[0027] Pre-process the dam height, reservoir capacity, porosity and the corresponding earth-rock dam burst peak flow to obtain historical earth-rock dam burst flow data;

[0028] Wherein, the preprocessing method is:

[0029] The input data and output data of the historical earth-rock dam burst flow data are normalized to the range of [0,1] through mapminmax. The dam height, reservoir capacity, porosity and the corresponding earth-rock dam burst peak flow at the same time point are taken as a sample. The dam height, reservoir capacity and porosity are taken as input data, and the earth-rock dam burst peak flow corresponding to the dam height, reservoir capacity and porosity is taken as output data.

[0030] Preferably, the method of performing performance evaluation and verification on the trained support vector machine model until it meets a preset performance evaluation index threshold is:

[0031] Use the training set to train the optimized support vector machine model;

[0032] During the training process, the support vector machine model uses the optimal C and γ values ​​to optimize the model by calculating the loss function to minimize the prediction error until it meets the preset performance evaluation index threshold;

[0033] Among them, prediction and error assessment include:

[0034] Use the trained support vector machine model to predict the test set data;

[0035] The root mean square error, coefficient of determination, and mean absolute error of the support vector machine model were calculated as performance evaluation indicators.

[0036] Preferably, the method of collecting the current earth-rock dam burst flow characteristic data and inputting the current earth-rock dam burst flow characteristic data into the trained support vector machine model, that is, obtaining the current predicted burst peak flow, is:

[0037] Collect current earth-rock dam breach flow characteristics data;

[0038] And input the current earth-rock dam burst flow characteristic data into the trained support vector machine model;

[0039] The trained support vector machine model can output the currently predicted peak flow.

[0040] Preferably, the specific process of evaluating the fitness value of a particle is: defining a fitness function to evaluate each particle, and calculating the fitness value of each particle, wherein the fitness function is: , where represents the fitness value of the i-th particle, represents the accuracy of the support vector machine model, represents the complexity of the support vector machine model, and They are and The weight of and The values ​​are 0.8 and 0.2 respectively, where , represents the position value of the jth particle in the population in the first dimension, Represents the position value of the jth particle in the population in the second dimension.

[0041] Preferably, the updated position formula is:

[0042] ;

[0043] In the formula, represents the original velocity vector, The inertia weight of the influence on the new vector, represents the individual acceleration constant, represents the social acceleration constant, represents a random function in the range [0,1]; represents the current position of the particle; pbest represents the individual optimal position; gbest represents the global optimal position.

[0044] The technical effects and advantages of the earth-rock dam burst flow prediction method based on data enhancement and adaptive network of the present invention are as follows:

[0045] By using the generative adversarial network model and attention mechanism to enhance the training set, and inputting the enhanced training set into the adaptive particle swarm optimization support vector machine model, accurate prediction of the peak flow of earth-rock dam breach can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the earth-rock dam burst flow prediction method based on data enhancement and adaptive network of the present invention;

[0047] Figure 2 It is a structural schematic diagram of the present invention;

[0048] Figure 3 It is a schematic diagram comparing the peak flow prediction results of the PSO-SVM model and the adaptive PSO-SVM model in the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Example 1

[0051] See also Figures 1 to 3 As shown, the earth-rock dam burst flow prediction method based on data enhancement and adaptive network described in this embodiment includes the following steps:

[0052] S1. Data collection: Collect and pre-process the data of historical earth-rock dam burst flow, and then obtain the historical earth-rock dam burst flow data;

[0053] S2. Generate synthetic data using GAN: Design and train a generative adversarial network to expand historical earth-rock dam burst flow data through the generative adversarial network, and then obtain an earth-rock dam burst flow data set;

[0054] S3. Feature extraction using the Attention mechanism: Use the Attention mechanism to extract features from the data in the earth-rock dam burst flow dataset, and weight the extracted features to obtain the earth-rock dam burst flow feature data, and divide the earth-rock dam burst flow feature data into a training set and a test set;

[0055] S4, constructing a support vector machine model, and using an adaptive particle swarm optimization algorithm to train the support vector machine model, thereby obtaining a trained support vector machine model;

[0056] S5. Perform performance evaluation and verification on the trained support vector machine model until it meets the preset performance evaluation index threshold;

[0057] S6. Collect the current earth-rock dam burst flow characteristic data, input the current earth-rock dam burst flow characteristic data into the trained support vector machine model, and obtain the current predicted burst peak flow.

[0058] Furthermore, the method of collecting and preprocessing the data of historical earth-rock dam burst flow rates to obtain the historical earth-rock dam burst flow rate data is as follows:

[0059] Collect historical data on earth-rock dam breach discharges, including dam height, reservoir capacity, porosity, and corresponding earth-rock dam breach peak discharges;

[0060] The dam height, reservoir capacity, porosity and corresponding earth-rock dam burst peak flow are preprocessed to obtain historical earth-rock dam burst flow data; wherein the preprocessing method is:

[0061] The input data and target data (output data) in the historical earth-rock dam burst flow data are normalized to [0,1] through mapminmax. The dam height, reservoir capacity, porosity and the corresponding earth-rock dam burst peak flow at the same time point are taken as a sample. The dam height, reservoir capacity and porosity are taken as input data, and the earth-rock dam burst peak flow corresponding to the dam height, reservoir capacity and porosity is taken as the target data (output data).

[0062] Specifically, 40 sets of actual data were collected from the earth-rock dam burst peak flow prediction task as historical earth-rock dam burst flow data.

[0063] Furthermore, the method of designing and training a generative adversarial network and expanding the historical earth-rock dam burst flow data by the generative adversarial network to obtain an earth-rock dam burst flow data set is as follows:

[0064] Design a generative adversarial network to generate earth-rock dam burst flow data similar to historical earth-rock dam burst flow data by training the generator, where the generator generates earth-rock dam burst flow data through random noise;

[0065] The discriminator determines whether the generated earth-rock dam burst flow data is real data, compares the generated earth-rock dam burst flow data with the historical earth-rock dam burst flow data, and then obtains the true and false probability value;

[0066] The generator continuously adjusts the generated earth-rock dam burst flow data through true and false probability values ​​to make it consistent with the historical earth-rock dam burst flow data;

[0067] The objective function of GAN is:

[0068] min G max D E x~pdata(x) [logD(x)] +E z~pz(z) [log(1-D(G(z)))];

[0069] In the formula, G(z) represents the generator, z represents random noise, x=G(z) represents the generated earth-rock dam burst flow data, D(x) represents the discriminator, which is used to output the probability that x belongs to the real data, pdata(x) represents the real data distribution, and pz(z) is the noise distribution of the generator input; E x~pdata(x) [logD(x)] represents the expectation of the discriminator’s judgment result on the historical earth-rock dam burst flow data x; E z~pz(z) [log(1-D(G(z)))] represents the expectation of the earth-rock dam burst flow data G(z) generated by the generator; min G The goal of the generator G is to minimize the discriminator's ability to identify; max D It means that the goal of the discriminator D is to maximize its ability to discriminate real data;

[0070] Through iterative training, the generated earth-rock dam outburst flow data are added to the historical earth-rock dam outburst flow data to obtain the earth-rock dam outburst flow data set.

[0071] Specifically, the generative adversarial network enhances the training data for burst flow prediction, helping the model to extract information from more diverse data and improving prediction accuracy. In addition to the GAN generative adversarial network, traditional data enhancement methods such as SMOTE (Synthetic Minority Over-sampling Technique) or enhancement methods based on simulated data generation can also be used. Although these methods can also effectively expand the training data, the diversity and quality of the generated synthetic data are lower than that of GAN.

[0072] It should be noted that by generating diverse training data samples through the generative adversarial network model and combining the attention mechanism to enhance the features of the data, the support vector machine model can better focus on the key features related to the burst flow prediction.

[0073] Furthermore, the Attention mechanism is used to extract features from the data in the earth-rock dam burst flow dataset, and the extracted features are weighted to obtain the characteristic data of the earth-rock dam burst flow. The earth-rock dam burst flow characteristic data is divided into a training set and a test set as follows:

[0074] The default input query matrix Q, key matrix K, and value matrix V are Q∈R nq×dk , K∈R nk×dk, V∈R nk×dv , where nq represents the number of queries, nk represents the number of keys, dk is the dimension of the query and key, dv is the dimension of the value, and R represents the prefix of the matrix dimension;

[0075] Calculate the attention score matrix Attention Scores by matrix multiplication, that is, Attention Scores=QK T , where K T is the transpose of the key matrix, and the attention score matrix has the size of nq×nk, which represents the similarity between each query and each key;

[0076] Scale the attention scores to obtain the scaled attention score matrix Scaled Scores, that is, Scaled Scores = QK T / √dk, where √dk is the square root of the bond dimension;

[0077] Apply the softmax function to each row in the scaled attention score matrix to obtain the attention weights Attention Weights, that is, Attention Weights=softmax(QK T / √dk);

[0078] Use the attention weight matrix and the value matrix to perform matrix multiplication to obtain the weighted summed output matrix Attention Output. The output matrix is ​​the characteristic data of earth-rock dam burst flow, that is, Attention Output = Attention Weights V;

[0079] The burst flow characteristic data of earth-rock dams are divided into training set and test set, with 80% of the burst flow characteristic data as the training set and 20% of the burst flow characteristic data as the test set.

[0080] Furthermore, a support vector machine model is constructed, and the support vector machine model is trained using an adaptive particle swarm optimization algorithm, and then the trained support vector machine model is obtained as follows:

[0081] The support vector machine model is constructed using the RBF (radial basis function) kernel;

[0082] The optimal parameters of the support vector machine model are obtained through the adaptive particle swarm optimization algorithm. The optimal parameters are the optimal solutions of the regularization parameter C and the kernel function parameter γ.

[0083] The support vector machine model is trained using the optimal parameters until the preset maximum number of evolutions or error threshold is reached, thus obtaining a trained support vector machine model.

[0084] Furthermore, the method of obtaining the optimal parameters of the support vector machine model through the adaptive particle swarm optimization algorithm is as follows:

[0085] Initialize the particle swarm: set the size of the particle swarm, initialize the position and speed of each particle. The position of the particle represents a set of parameters, namely the regularization parameter C and the kernel function parameter γ, where the speed represents the direction and rate of change of these values. Initialize the individual optimal position (pBest) and the global optimal position (gBest) of the particle. The individual optimal position is the best position (i.e. the best parameter combination) found by each particle during the search process, and the global optimal position is the best position found in the entire particle swarm.

[0086] Evaluate the fitness value of particles: define a fitness function to evaluate each particle and calculate the fitness value of each particle, where the fitness function is: , where represents the fitness value of the i-th particle, represents the accuracy of the support vector machine model, represents the complexity of the support vector machine model, and They are and The weight of and The values ​​can be 0.8 and 0.2 respectively, where , represents the position value of the jth particle in the population in the first dimension (or the first parameter), Represents the position value of the jth particle in the population in the second dimension (or the first parameter);

[0087] Update particle position and speed: Update the individual optimal position (pBest) of each particle according to the fitness value. If the fitness value of the current particle is better than the fitness value of the global optimal position (gBest), update the global optimal position (gBest). Introduce an adaptive mechanism to adjust the particle's speed and position update strategy, and use the updated position formula to calculate the new position and speed of each particle.

[0088] The update formula of position is:

[0089] ;

[0090] In the formula, represents the original velocity vector, The inertia weight of the influence on the new vector, represents the individual acceleration constant, represents the social acceleration constant, represents a random function in the range [0,1]; Indicates the current position of the particle;

[0091] The fitness value of the particle and the update of the particle position are repeatedly evaluated until a predetermined number of iterations is reached or other stopping conditions are met.

[0092] Specifically, in addition to the particle swarm optimization (PSO) algorithm, other optimization algorithms such as genetic algorithm (GA) and ant colony optimization (ACO) may also be used. Although these algorithms may produce similar effects in some cases, the convergence speed and search efficiency of the PSO algorithm are generally higher and more suitable for the needs of the present invention.

[0093] Furthermore, the performance of the trained support vector machine model is evaluated and verified until it meets the preset performance evaluation index threshold as follows:

[0094] Use the training set to train the optimized support vector machine model;

[0095] During the training process, the support vector machine model uses the optimal C and γ values ​​to optimize the model by calculating the loss function to minimize the prediction error;

[0096] Through the support vector regression algorithm, a support vector machine model is obtained to perform regression prediction on the earth-rock dam burst flow.

[0097] Prediction and Error Assessment:

[0098] Use the trained support vector machine model to predict the test set data;

[0099] Calculate the model's root mean square error (RMSE), coefficient of determination (R²), and mean absolute error (MAE) to evaluate the model's performance;

[0100] Specifically, the prediction results can also be denormalized to restore the data to the actual value range; Result analysis and visualization: Draw a comparison chart of the prediction results, compare the actual values ​​and predicted values ​​of the training set and the test set, and intuitively display the prediction performance of the model;

[0101] Furthermore, after actual verification, for the same enhanced data set (i.e., earth-rock dam burst flow characteristic data), the root mean square error (RMSE) of the adaptive PSO-SVM model (the support vector machine model in the present invention) on the test set is reduced by 22.86% compared with the traditional PSO-SVM model. The specific values ​​of RMSE are: adaptive PSO-SVM model, 0.0153; traditional PSO-SVM, 0.019834.

[0102] Furthermore, the method of collecting the current earth-rock dam burst flow characteristic data and inputting the current earth-rock dam burst flow characteristic data into the trained support vector machine model to obtain the current predicted burst peak flow is:

[0103] Collect current earth-rock dam breach flow characteristics data;

[0104] And input the current earth-rock dam burst flow characteristic data into the trained support vector machine model;

[0105] The trained support vector machine model can output the currently predicted peak flow.

[0106] In this embodiment, the present invention introduces a generative adversarial network to perform data enhancement and generate more data, which can help the support vector machine model to obtain better training results when data is scarce; by combining the attention mechanism to automatically focus on important features in the data, the expression ability and prediction performance of the support vector machine model are improved; by adjusting the parameters of the support vector machine model through an adaptive particle swarm optimization algorithm, the generalization ability of the support vector machine model can be enhanced, so that it can better adapt to different earth-rock dam breach scenarios; and then through data enhancement and feature adaptive adjustment, the prediction accuracy and stability of the model are significantly improved, and the problems of data scarcity, insufficient feature extraction and poor model generalization ability in the prior art are solved, so that it has broad application prospects in the field of earth-rock dam breach flow prediction;

[0107] In addition, in this embodiment, 40 groups of earth-rock dam breach cases are obtained through independent experiments, and the historical earth-rock dam breach flow data are expanded to 200 groups through data enhancement (generative adversarial network); the PSO-SVM model and the adaptive PSO-SVM model are used to predict the peak breach flow respectively, and the analysis involves four parameters: dam height, reservoir capacity, porosity and peak breach flow; among them, dam height, reservoir capacity and porosity are used as input parameters, and peak breach flow is used as output parameter; the data set (i.e., earth-rock dam breach flow characteristic data) is divided into training set and test set according to the ratio of 8:2, and the prediction results are as follows: Figure 3 As shown in the figure; for the same enhanced data set (i.e., earth-rock dam burst flow characteristic data), the adaptive PSO-SVM model (the support vector machine model in the present invention) has a root mean square error (RMSE) of 21.42% on the test set compared with the traditional PSO-SVM model; specifically, the RMSE of the adaptive PSO-SVM model is 0.015085, while the RMSE of the traditional PSO-SVM is 0.019196.

[0108] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0109] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0110] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0111] Finally: 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 in the protection scope of the present invention.

Claims

1. A method for predicting earth-rock dam burst flow based on data enhancement and adaptive network, characterized in that: The following steps are involved: S1. Collect and pre-process historical earth-rock dam burst flow data, and then obtain historical earth-rock dam burst flow data; S2. Design and train a generative adversarial network, and expand the historical earth-rock dam burst flow data through the generative adversarial network to obtain an earth-rock dam burst flow data set; the specific method is: design a generative adversarial network, and generate earth-rock dam burst flow data similar to the historical earth-rock dam burst flow data in the training set by training the generator, wherein the generator generates earth-rock dam burst flow data through random noise; the discriminator determines whether the generated earth-rock dam burst flow data is real data, and compares the generated earth-rock dam burst flow data with the historical earth-rock dam burst flow data to obtain the true and false probability values; the generator continuously adjusts the generated earth-rock dam burst flow data through the true and false probability values ​​to make it consistent with the historical earth-rock dam burst flow data; the objective function of GAN is: min G max D E x~pdata(x) [logD(x)] +E z~pz(z) [log(1-D(G(z)))]; In the formula, G(z) represents the generator, z represents random noise, x=G(z) represents the generated earth-rock dam burst flow data, D(x) represents the discriminator, which is used to output the probability that x belongs to the real data, pdata(x) represents the real data distribution, and pz(z) is the noise distribution of the generator input; E x~pdata(x) [logD(x)] represents the expectation of the discriminator’s judgment result on the historical earth-rock dam burst flow data x; E z~pz(z) [log(1-D(G(z)))] represents the expectation of the earth-rock dam burst flow data G(z) generated by the generator; min G The goal of the generator G is to minimize the discriminator's ability to identify; max D It means that the goal of the discriminator D is to maximize its ability to discriminate real data. Through iterative training, the generated earth-rock dam burst flow data is added to the historical earth-rock dam burst flow data to obtain the earth-rock dam burst flow data set. S3. Use the Attention mechanism to extract features from the data in the earth-rock dam burst flow dataset, and weight the extracted features to obtain the characteristic data of the earth-rock dam burst flow, and divide the earth-rock dam burst flow characteristic data into a training set and a test set; the specific method is: the preset input query matrix Q, key matrix K, and value matrix V are Q∈R nq×dk , K∈R nk×dk , V∈R nk×dv , where nq represents the number of queries, nk represents the number of keys, dk is the dimension of the query and key, dv is the dimension of the value, and R represents the prefix of the matrix dimension; Calculate the attention score matrix Attention Scores by matrix multiplication, that is, Attention Scores=QK T , where K T is the transpose of the key matrix, and the attention score matrix has the size of nq×nk, which represents the similarity between each query and each key; Scale the attention scores to obtain the scaled attention score matrix Scaled Scores, that is, ScaledScores=QK T / √dk, where √dk is the square root of the bond dimension; Apply the softmax function to each row in the scaled attention score matrix to obtain the attention weights Attention Weights, that is, Attention Weights=softmax(QK T / √dk); Use the attention weight matrix and the value matrix to perform matrix multiplication to obtain the weighted summed output matrix AttentionOutput. The output matrix is ​​the characteristic data of earth-rock dam burst flow, that is, Attention Output = AttentionWeights V; The earth-rock dam burst flow characteristic data are divided into training set and test set, with 80% of the earth-rock dam burst flow characteristic data as the training set and 20% of the earth-rock dam burst flow characteristic data as the test set; S4, constructing a support vector machine model, and using an adaptive particle swarm optimization algorithm to train the support vector machine model, thereby obtaining a trained support vector machine model; the specific method is: using an RBF kernel to construct a support vector machine model; obtaining the optimal parameters of the support vector machine model through an adaptive particle swarm optimization algorithm, the optimal parameters being the optimal solution of the regularization parameter C and the kernel function parameter γ; using the optimal parameters to train the support vector machine model until a preset maximum number of evolutions or an error threshold is reached, thereby obtaining a trained support vector machine model; The method of obtaining the optimal parameters of the support vector machine model by using the adaptive particle swarm optimization algorithm is: Initialize the particle swarm: set the size of the particle swarm, initialize the position and velocity of each particle, and the position of the particle represents a set of parameters, namely the regularization parameter C and the kernel function parameter γ; Evaluate the fitness value of particles: define a fitness function to evaluate each particle and calculate the fitness value of each particle; Update particle position and speed: Update the individual optimal position of each particle according to the fitness value. If the fitness value of the current particle is better than the fitness value of the global optimal position, update the global optimal position, introduce an adaptive mechanism to adjust the particle's speed and position update strategy, and use the updated position formula to calculate the new position and speed of each particle; Repeatedly evaluate the fitness value of the particle and update the particle position until the predetermined number of iterations is reached or other stopping conditions are met; S5. Perform performance evaluation and verification on the trained support vector machine model until it meets the preset performance evaluation index threshold; S6. Collect the current earth-rock dam burst flow characteristic data, input the current earth-rock dam burst flow characteristic data into the trained support vector machine model, and obtain the current predicted burst peak flow.

2. The earth-rock dam burst flow prediction method based on data enhancement and adaptive network according to claim 1 is characterized in that: The method of collecting and preprocessing the data of historical earth-rock dam burst flow rates, and then obtaining the historical earth-rock dam burst flow rate data is as follows: Collect historical data on earth-rock dam breach discharges, including dam height, reservoir capacity, porosity, and corresponding earth-rock dam breach peak discharges; Pre-process the dam height, reservoir capacity, porosity and the corresponding earth-rock dam burst peak flow to obtain historical earth-rock dam burst flow data; Wherein, the preprocessing method is: The input data and output data of the historical earth-rock dam burst flow data are normalized to the range of [0,1] through mapminmax. The dam height, reservoir capacity, porosity and the corresponding earth-rock dam burst peak flow at the same time point are taken as a sample. The dam height, reservoir capacity and porosity are taken as input data, and the earth-rock dam burst peak flow corresponding to the dam height, reservoir capacity and porosity is taken as output data.

3. The method for predicting earth-rock dam burst flow based on data enhancement and adaptive network according to claim 1 is characterized in that: The method of performing performance evaluation and verification on the trained support vector machine model until it meets the preset performance evaluation index threshold is as follows: Use the training set to train the optimized support vector machine model; During the training process, the support vector machine model uses the optimal C and γ values ​​to optimize the model by calculating the loss function to minimize the prediction error until it meets the preset performance evaluation index threshold; Among them, prediction and error assessment include: Use the trained support vector machine model to predict the test set data; The root mean square error, coefficient of determination, and mean absolute error of the support vector machine model were calculated as performance evaluation indicators.

4. The method for predicting earth-rock dam burst flow based on data enhancement and adaptive network according to claim 1 is characterized in that: The method of collecting the current earth-rock dam burst flow characteristic data and inputting the current earth-rock dam burst flow characteristic data into the trained support vector machine model, that is, obtaining the current predicted burst peak flow, is as follows: Collect current earth-rock dam breach flow characteristics data; And input the current earth-rock dam burst flow characteristic data into the trained support vector machine model; The trained support vector machine model can output the currently predicted peak flow.

5. The method for predicting earth-rock dam burst flow based on data enhancement and adaptive network according to claim 1 is characterized in that: The specific process of evaluating the fitness value of a particle is: define a fitness function to evaluate each particle and calculate the fitness value of each particle, where the fitness function is: , where represents the fitness value of the i-th particle, represents the accuracy of the support vector machine model, represents the complexity of the support vector machine model, and They are and The weight of and The values ​​are 0.8 and 0.2 respectively, where , represents the position value of the jth particle in the population in the first dimension, Represents the position value of the jth particle in the population in the second dimension.

6. The method for predicting earth-rock dam burst flow based on data enhancement and adaptive network according to claim 1, characterized in that: The updated position formula is: ； In the formula, represents the original velocity vector, The inertia weight of the influence on the new vector, represents the individual acceleration constant, represents the social acceleration constant, represents a random function in the range [0,1]; represents the current position of the particle; pbest represents the individual optimal position; gbest represents the global optimal position.

Images