Earth and rockfill dam burst peak flow prediction method based on elite strategy

Abstract

The invention belongs to the technical field of machine learning, and discloses an earth and rockfill dam outburst peak flow prediction method based on an elite strategy, and the method comprises the steps: obtaining historical earth and rockfill dam outburst feature data and a corresponding outburst peak flow to construct an earth and rockfill dam outburst data set; constructing a BP neural network, and improving the genetic algorithm by adopting an elitist strategy in the selection, intersection and variation processes of the genetic algorithm to obtain an improved genetic algorithm; taking the weight and bias of the BP neural network as individuals of the improved genetic algorithm, and optimizing the BP neural network by using the improved genetic algorithm to obtain an improved GA-BP neural network based on an elitist strategy; inputting the data in the data set into a GA-BP neural network to obtain a predicted outburst peak flow; based on an elitist strategy improved GA-BP neural network model, the accuracy and real-time performance of dam break peak flow prediction are greatly improved, and timely and effective decision support is provided for disaster prevention and reduction and emergency response; and the common local optimal solution difficulty in the traditional GA is also avoided.

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 peak flow based on an elite strategy. Background Art

[0002] In the field of flood peak flow prediction, existing technologies mainly rely on two methods: numerical simulation and experimental model testing. Numerical simulation technology estimates peak flow by modeling rainfall-runoff processes and river flow processes. Testing methods based on experimental models are also widely used in dam break and flood prediction research. Such methods estimate peak flow by experimentally analyzing the dam break process and combining empirical formulas. In the field of neural network hydrological prediction, BP neural network (back propagation neural network) is widely used to deal with complex nonlinear relationships, especially for predicting water level or sediment volume downstream of floods.

[0003] However, the existing traditional BP neural network lacks a global optimization strategy and cannot effectively solve the complexity and nonlinear problems in the prediction of earth-rock dam burst flow, which easily affects the stability and accuracy of the prediction results; in the evolutionary process, excellent individuals may not be fully retained and transmitted, resulting in the optimization process failing to effectively explore the entire search space; and thus the existing earth-rock dam burst prediction method lacks sufficient accuracy and real-time performance, making it inconvenient to provide timely and effective decision support for emergency response.

[0004] In view of this, the present invention proposes a method for predicting peak flow of earth-rock dam breach based on elite strategy to solve the above problems. Summary of the invention

[0005] In order to overcome the above defects of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a method for predicting the peak flow of earth-rock dam burst based on elite strategy, comprising the following steps:

[0006] S1. Acquire historical earth-rock dam breach characteristic data and corresponding earth-rock dam breach peak flow, and construct an earth-rock dam breach dataset using the historical earth-rock dam breach characteristic data and corresponding earth-rock dam breach peak flow;

[0007] S2. Constructing a BP neural network model: dividing the historical earth-rock dam burst data set into a training set and a test set in a ratio of 8:2, constructing the historical earth-rock dam burst feature data into a feature vector P, wherein each historical earth-rock dam burst feature data and its corresponding earth-rock dam burst peak flow are taken as a sample; using the newff function to construct a BP neural network model, wherein the input layer is the feature vector P, and the output layer is the corresponding earth-rock dam burst peak flow; during the network training process, the input data is passed to the hidden layer through the associated weights and biases, and after being processed by the activation function, the output layer is used to give the predicted earth-rock dam burst peak flow;

[0008] S3. Train and optimize the BP neural network model through the historical earth-rock dam breach data set and genetic algorithm: train the BP neural network model through the train function, and use the back propagation algorithm to optimize the weights and biases of the network during the training process to minimize the mean square error until the preset cutoff condition is reached; then, optimize the weights and biases of the BP neural network model through the genetic algorithm, and assign the weights and biases optimized by the genetic algorithm to the BP neural network model;

[0009] The method of optimizing the weights and biases of the BP neural network model by genetic algorithm is as follows:

[0010] Initialize the population, where each individual represents the weight and bias of a network parameter of the BP neural network model, and a uniform distribution is used for individual initialization;

[0011] Define the fitness function, its formula is:

[0012] ;

[0013] In the formula, is the fitness value, is the number of samples in the training set, is the serial number of each sample in the training set, is the predicted peak flow of earth-rock dam breach, is the peak flow of earth-rock dam burst corresponding to the samples in the training set; individuals with higher fitness values ​​are selected to enter the next generation population through the normalized geometric selection method, and the best individuals are retained in each generation by the elite strategy; arithmetic crossover is selected as the crossover method, the genes of the two parent individuals are combined to generate new offspring individuals, and the crossover operation uses the weights and biases of the two parents to reorganize and generate new individuals; then the weights and biases of the individuals are mutated by the non-uniform mutation method, and the mutation includes small-scale random changes; then, in each iteration of the mutation operation, if the fitness value of the mutated individual is the same as that of the current best individual, the individual is kept as the optimal solution until the preset iteration conditions are met and the optimal solution is found, and this optimal solution is used as the weight and bias of the BP neural network model;

[0014] The crossover operation formula is:

[0015] ;

[0016] ;

[0017] In the formula, is the cross coefficient, and are the genes of the two parent individuals, and They are two offspring individuals generated after the crossover of two parent individuals;

[0018] The operational formula of the mutation is:

[0019] ;

[0020] In the formula, For the individual after mutation, is the individual before mutation, δ is the randomly generated mutation amount;

[0021] Afterwards, the trained GA-BP neural network model based on the elite strategy improvement is obtained;

[0022] S4. Obtain the current earth-rock dam burst characteristic data, input the current earth-rock dam burst characteristic data into the trained GA-BP neural network model based on the elite strategy improvement, and obtain the predicted earth-rock dam burst peak flow.

[0023] Preferably, the historical earth-rock dam breach characteristic data include dam height, dam top width, dam bottom width, dam width, initial water depth, median particle size, whether slots are opened, reservoir capacity and porosity.

[0024] Preferably, before constructing the BP neural network model, the data in the historical earth-rock dam failure characteristic data are normalized in sequence.

[0025] Preferably, in S4, the method of inputting the current earth-rock dam breach characteristic data into the trained GA-BP neural network model based on the elite strategy improvement to obtain the predicted earth-rock dam breach peak flow is:

[0026] Collect current earth-rock dam breach characteristic data and normalize the current earth-rock dam breach characteristic data;

[0027] The normalized current earth-rock dam breach characteristic data is input into the trained GA-BP neural network model based on the elite strategy improvement;

[0028] Furthermore, the trained GA-BP neural network model based on the elite strategy improvement can output the predicted peak flow of earth-rock dam breach.

[0029] Preferably, the specific process of transmitting the historical earth-rock dam failure characteristic data in the input layer to the hidden layer through the associated weights and biases, and the output layer giving the predicted earth-rock dam failure peak flow is:

[0030] Select the logsig formula as the activation function. The activation function is as follows:

[0031] ;

[0032] In the formula, is the input value, is the base of natural logarithms;

[0033] The data passed to the hidden layer is the weighted and biased data of the input layer. Assume that the input layer has n nodes and the hidden layer has m nodes. The input data is , , the weight is , the bias is ,in, is the weight from the jth input layer node to the ith hidden layer node, is the weight from the jth input layer node to the i+1th hidden layer node, is the bias of the i-th hidden layer node; then, the input of the hidden layer node is the weighted sum of the i-th hidden layer node, expressed as the following formula:

[0034] ;

[0035] Finally, the output layer is calculated based on the output of the hidden layer to generate the predicted earth-rock dam breach peak flow rate, which is obtained by the following formula:

[0036] ;

[0037] In the formula, is the weight from the mth hidden layer node to the output layer, is the predicted peak flow of earth-rock dam failure, represents the bias of the jth node in the hidden layer, is the bias of the output layer.

[0038] The technical effects and advantages of the method for predicting peak flow of earth-rock dam burst based on elite strategy of the present invention are as follows:

[0039] By adopting an improved genetic algorithm with an elite retention strategy, the stability and search efficiency of the optimization process are enhanced, and the local optimal solution dilemma common in traditional GA is effectively avoided; by integrating the BP neural network model optimized by GA, the accuracy and real-time performance of dam break peak flow prediction are greatly improved, which can provide timely and effective decision-making support for disaster prevention, mitigation and emergency response; and the dam break peak flow prediction method provided by the present invention is not only applicable to different types of dams and river basins, but also has good adaptability and can be used in combination with other hydrological models and meteorological data sources to further enhance the prediction capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of the method for predicting the peak flow rate of earth-rock dam burst based on the elite strategy of the present invention;

[0041] Figure 2 It is a structural schematic diagram of the BP neural network model in the present invention;

[0042] Figure 3 It is a specific schematic diagram of the present invention. DETAILED DESCRIPTION

[0043] 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.

[0044] Example 1

[0045] See also Figure 1 and Figure 2 , Figure 3 As shown, the method for predicting the peak flow rate of earth-rock dam breach based on the elite strategy described in this embodiment includes the following steps:

[0046] S1, obtaining historical earth-rock dam failure characteristic data and earth-rock dam failure peak flow corresponding to the historical earth-rock dam failure characteristic data, and then obtaining a historical earth-rock dam failure data set;

[0047] S2. Constructing a BP neural network model: dividing the historical earth-rock dam burst data set into a training set and a test set in a ratio of 8:2, constructing the historical earth-rock dam burst feature data into a feature vector P, wherein each historical earth-rock dam burst feature data and its corresponding earth-rock dam burst peak flow are taken as a sample; using the newff function to construct a BP neural network model, wherein the input layer is the feature vector P, and the output layer is the corresponding earth-rock dam burst peak flow; during the network training process, the input data is passed to the hidden layer through the associated weights and biases, and after being processed by the activation function, the output layer is used to give the predicted earth-rock dam burst peak flow;

[0048] S3. Train and optimize the BP neural network model through the historical earth-rock dam breach data set and genetic algorithm: train the BP neural network model through the train function, and use the back propagation algorithm to optimize the weights and biases of the network during the training process to minimize the mean square error until the preset cutoff condition is reached; then, optimize the weights and biases of the BP neural network model through the genetic algorithm, and assign the weights and biases optimized by the genetic algorithm to the BP neural network model;

[0049] The method of optimizing the weights and biases of the BP neural network model by genetic algorithm is as follows:

[0050] Initialize the population, where each individual represents the weight and bias of a network parameter of the BP neural network model, and a uniform distribution is used for individual initialization;

[0051] Define the fitness function, its formula is:

[0052] ,

[0053] In the formula, is the fitness value, is the number of samples in the training set, is the serial number of each sample in the training set, is the predicted peak flow of earth-rock dam breach, is the peak flow of earth-rock dam burst corresponding to the samples in the training set; individuals with higher fitness values ​​are selected to enter the next generation population through the normalized geometric selection method, and the best individuals are retained in each generation by the elite strategy; arithmetic crossover is selected as the crossover method, the genes of the two parent individuals are combined to generate new offspring individuals, and the crossover operation uses the weights and biases of the two parents to reorganize and generate new individuals; then the weights and biases of the individuals are mutated by the non-uniform mutation method, and the mutation includes small-scale random changes; then, in each iteration of the mutation operation, if the fitness value of the mutated individual is the same as that of the current best individual, the individual is kept as the optimal solution until the preset iteration conditions are met and the optimal solution is found, and this optimal solution is used as the weight and bias of the BP neural network model;

[0054] The crossover operation formula is:

[0055] ;

[0056] ;

[0057] In the formula, is the cross coefficient, and are the genes of the two parent individuals, and They are two offspring individuals generated after the crossover of two parent individuals;

[0058] The operational formula of the mutation is:

[0059] ;

[0060] In the formula, For the individual after mutation, is the individual before mutation, δ is the randomly generated mutation amount;

[0061] Afterwards, the trained GA-BP neural network model based on the elite strategy improvement is obtained;

[0062] S4. Obtain the current earth-rock dam burst characteristic data, input the current earth-rock dam burst characteristic data into the trained GA-BP neural network model based on the elite strategy improvement, and obtain the predicted earth-rock dam burst peak flow.

[0063] Furthermore, historical earth-rock dam failure characteristic data include dam height H d , dam crest width L c , dam bottom width L b , dam width L d (Dam width L d is the width of the dam perpendicular to the direction of water flow), initial water depth H, median particle size D50 , whether to slot (yes and no can be represented by 1 and 0 respectively), storage capacity S c and porosity P.

[0064] Furthermore, before constructing the BP neural network model, the data in the historical earth-rock dam breach characteristic data are normalized in turn (the normalization is performed using the formula of the minimum-maximum normalization method).

[0065] The way to construct the BP neural network model is:

[0066] The historical earth-rock dam breach dataset is divided into a training set and a test set in a ratio of 8:2, and the historical earth-rock dam breach feature data are used to construct a feature vector P, where each historical earth-rock dam breach feature data in the historical earth-rock dam breach dataset and its corresponding earth-rock dam breach peak flow are taken as a sample;

[0067] The BP neural network model is constructed using the newff function. The characteristic vector P constructed from the historical earth-rock dam breach characteristic data is used as the input layer, and the corresponding earth-rock dam breach peak flow is used as the output layer. The number of nodes in the input layer is the number of data in the historical earth-rock dam breach characteristic data, and the hidden layer is set to 5 hidden layer nodes.

[0068] The historical earth-rock dam breach characteristic data in the input layer are transmitted to the hidden layer through the associated weights and biases, and the output layer gives the predicted earth-rock dam breach peak flow;

[0069] Specifically, the logsig formula is selected as the activation function, and the activation function is as follows:

[0070] ;

[0071] In the formula, is the input value (usually the weighted sum of neurons, including bias terms), is the base of natural logarithms (approximately equal to 2.71828);

[0072] The data input to the hidden layer is the weighted and biased data of the input layer. Assuming that the input layer has n nodes and the hidden layer has m nodes, the input data (historical earth-rock dam breach characteristic data) is , , the weight is , the bias is (in, is the weight from the jth input layer node to the ith hidden layer node, is the weight from the jth input layer node to the i+1th hidden layer node, is the bias of the i-th hidden layer node), then the input of the hidden layer node (that is, the weighted sum of the i-th hidden layer node) is expressed as the following formula:

[0073] ;

[0074] Finally, the output layer is calculated based on the output of the hidden layer to generate the predicted earth-rock dam breach peak flow rate, which is obtained by the following formula:

[0075] ;

[0076] In the formula, is the weight from the mth hidden layer node to the output layer, is the predicted peak flow of earth-rock dam failure, represents the bias of the jth node in the hidden layer, is the bias of the output layer.

[0077] For example, Figure 2 As shown, preset ( ) is the input data, the number of hidden layer nodes is 5, the number of output layer nodes is 1, and the network model is constructed using the newff function; the data in the input layer is associated with the weight ( ) and bias The data is transmitted to the hidden layer, and the output layer gives the predicted earth-rock dam burst peak flow rate. When the GA-BP neural network model improved based on the elite strategy is initialized, the input data (historical earth-rock dam burst characteristic data) and the target data (earth-rock dam burst peak flow rate corresponding to the historical earth-rock dam burst characteristic data) are trained through the back propagation algorithm. Then, the network weights are optimized through the back propagation algorithm to enable it to adapt to complex nonlinear relationships, and the mapping relationship between the dam burst peak flow rate and the input characteristics is preliminarily established.

[0078] It should be noted that the dataset has 40 sets of data, including 9 parameters, and the ratio of training set to test set is 8:2; the experimental data is imported from Excel files and the data is normalized; the dataset includes input features and target outputs, the input features are 9 dimensions, and the target output is 1 dimension; the dataset is randomly divided into training sets (32 sets of data) and test sets (8 sets of data), and normalized by the mapminmax function to ensure that the range of input and output data is between [0,1]; thus, the training set is used to learn the patterns and regularities of the data, and the test set is used to check the accuracy of the trained model.

[0079] Furthermore, the BP neural network model is trained and optimized through the historical earth-rock dam breach data set and genetic algorithm, and the trained GA-BP neural network model based on the elite strategy improvement is obtained as follows:

[0080] The BP neural network model is trained through the train function, and during the training process, the back propagation algorithm is used to optimize the weights and biases of the network to minimize the mean square error until the preset cutoff condition is reached;

[0081] Subsequently, the weights and biases of the BP neural network model are optimized by genetic algorithm, and the weights and biases optimized by genetic algorithm are assigned to the BP neural network model, so that the trained GA-BP neural network model based on elite strategy improvement can be obtained;

[0082] Specifically, the training cutoff condition of the genetic algorithm is a maximum of 50 generations, and the training cutoff condition of the BP neural network model is a maximum of 1000 iterations or a mean square error less than 1e-6; the cutoff condition of the GA-BP neural network model is to complete the training of the genetic algorithm first, and then continue to optimize through the training of the BP neural network until the cutoff condition of the BP neural network model is met.

[0083] Furthermore, the weights and biases of the BP neural network model are optimized by genetic algorithm as follows:

[0084] Initialize the population, where each individual represents the weight and bias (i.e., possible solution) of a network of the BP neural network model. Use uniform distribution for individual initialization to ensure that the entire solution space is fully explored; as shown in Table 1 (the explanation of the parameters in Table 1 (the selection of each value is the optimal value after repeated experimental verification));

[0085] Table 1 Parameters of genetic algorithm parameter value Genetic Algebra 50 Population size 5 Selecting initial function parameters 0.09 Crossover function initial parameters 2 Initial parameters of the variogram [2 gen 3]

[0086] Specifically, population initialization is the first step in generating candidate solutions. The initialization process generates an initial population based on the set variable boundaries and encoding methods, sets the parameter boundaries to [-1,1], and then generates the solution vector of the initial individual based on these boundaries. The population size is pop_num, and the dimension of each individual is the number of optimized parameters S (i.e., the weights and biases of the BP neural network model):

[0087] The calculation formula is: population = rand (pop_num, S) × (bmax-bmin) + bmin;

[0088] Among them, bmax and bmin are the upper and lower limits of the parameter respectively, and rand generates random numbers between [0,1];

[0089] The fitness function gabpEval is defined to evaluate the performance of each individual and also adjust some parameter values ​​in iteration. The fitness value can be measured by the fitness function (improved root mean square error (RMSE)) on the training set, and its formula is: ;

[0090] In the formula, is the fitness value, is the number of samples in the training set, is the serial number of each sample in the training set, is the predicted peak flow of earth-rock dam breach, is the peak flow rate of earth-rock dam breach corresponding to the samples in the training set;

[0091] Specifically, this is a fitness function based on the root mean square error (RMSE). Since the goal of a genetic algorithm is usually to minimize the error, the fitness value is inversely proportional to the error here, which means that the higher the fitness value, the smaller the prediction error of the model. The genetic algorithm will tend to select individuals with larger fitness values ​​(i.e., individuals with smaller errors) for the next round of evolution.

[0092] The "normalized geometric selection (normGeomSelect)" method is used to select individuals with higher fitness values ​​to enter the next generation of population. The elite strategy is used to ensure that the best individuals (that is, the geometric selection method gives high fitness individuals a greater selection probability, while ensuring that the individuals with the highest fitness are definitely retained) are retained in each generation to prevent excellent solutions from being destroyed by genetic operations. This is done by ensuring that the best individuals are directly passed on to the next generation. Therefore, the elite strategy speeds up the search process and enhances the stability of the algorithm by ensuring that the best individuals are not lost (that is, in the code, the best of each generation is placed at the top of the new set to avoid loss in subsequent operations);

[0093] Select "arithmetic crossover" as the crossover method to combine the genes of two parent individuals to generate new offspring individuals; the crossover operation uses the weights and bias parts of the two parents to recombine and generate new individuals to enhance the diversity of solutions;

[0094] ;

[0095] ;

[0096] In the formula, is the cross coefficient, which is a random weight coefficient between the interval [0,1] and is usually generated by a random number; and are the genes of the two parent individuals, and They are two offspring individuals generated after the crossover of two parent individuals;

[0097] In each iteration, the code implements the crossover operation in the following way:

[0098] Specifically, the code is as follows:

[0099] 【% If the individual after crossover becomes the best individual, remain unchanged

[0100] if isequal(c1, best)

[0101] c1 = best

[0102] end

[0103] if isequal(c2, best)

[0104] c2 = best:

[0105] end

[0106] endPop(a,:) = c1;

[0107] endPop(b,:)=c2;]

[0108] In this code, c1 and c2 are two offspring individuals after crossover; the code checks whether these two offspring are the same as the current best individual (best), and if they are the same, they are kept as the best individuals to ensure that the best individual is not destroyed;

[0109] The non-uniform mutation method is used to increase the diversity of the population and avoid premature convergence by making small random changes to the weights and biases of individuals.

[0110] The mutation operation formula is:

[0111] In the formula, For the individual after mutation, is the individual before mutation, δ is the randomly generated mutation amount;

[0112] In each iteration, the code implements the crossover operation in the following way:

[0113] 【% If the individual after mutation is the best individual, then remain unchanged

[0114] if isequal(c1, best)

[0115] c1 = best:

[0116] end

[0117] endPop(a,:)= c1; % Put the mutated individuals back into the population].

[0118] Furthermore, in each iteration of the mutation operation, if the fitness value of the mutated individual is the same as that of the current best individual, the individual will be kept as the optimal solution until the maximum number of iterations or the change in fitness value is lower than a certain threshold, and this optimal solution will be used as the weight and bias of the BP neural network model;

[0119] For example, in a mutation operation, if the mutated individual is the same as the current best individual (best), the code will keep the individual as the optimal solution, thus ensuring the effectiveness of the elite strategy;

[0120] Specifically, in each iteration, the code implements the elite strategy in the following way:

[0121] [[bval,bindx]=max(startPop(:,xzomeLength));

[0122] % Get the fitness value and index of the best individual in the current population

[0123] best = startPop(bindx,:);%Record the current best individual]

[0124] In this part, the code finds the best individual from the current population. Next, the elite individual will be retained in the next generation of the population.

[0125] 【% Ensure that the best individuals are retained to the next generation,

[0126] endPop(1,:)= best; %Put the best individual in the first position of the next generation]

[0127] Here, the best individual (best) is directly placed in the first position of the next generation population (endPop), thus ensuring that it will not be lost in subsequent crossover or mutation processes.

[0128] Among them, after the BP neural network model training is completed, the training set and the test set are used for simulation to generate prediction results. The simulation results are obtained through the sim function and restored to the actual value (the peak flow of earth-rock dam burst corresponding to the historical earth-rock dam burst characteristic data) through the denormalization process. The prediction effect of the model (the predicted peak flow of earth-rock dam burst) is evaluated through indicators such as RMSE (root mean square error), R² (coefficient of determination), and MAE (mean absolute error).

[0129] Furthermore, by inputting the current earth-rock dam breach characteristic data into the trained GA-BP neural network model based on the elite strategy improvement, the predicted earth-rock dam breach peak flow can be obtained as follows:

[0130] Collect current earth-rock dam breach characteristic data and normalize the current earth-rock dam breach characteristic data;

[0131] The normalized current earth-rock dam breach characteristic data is input into the trained GA-BP neural network model based on the elite strategy improvement;

[0132] Then, the trained GA-BP neural network model based on the elite strategy can output the predicted peak flow of earth-rock dam breach;

[0133] Among them, for the same data set (historical earth-rock dam breach data set), the root mean square error (RMSE) of the test set of the GA-BP neural network improved based on the elite strategy was reduced by 63.47% and 13.99% respectively compared with the BP network and GA-BP neural network. The specific values ​​of RMSE are: improved GA-BP, 0.0062234; GA-BP, 0.00724; BP, 0.017047.

[0134] In this embodiment, by combining the global optimization capability of the genetic algorithm with the nonlinear modeling advantages of the BP neural network, the local optimal solution problem of the traditional BP neural network is effectively overcome, and the global search capability of the algorithm in the evolutionary process is enhanced through an improved selection operator, thereby significantly improving the accuracy and stability of the dam break peak flow prediction.

[0135] 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.

[0136] 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.

[0137] 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.

[0138] 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 peak flow of earth-rock dam burst based on elite strategy, characterized in that: The following steps are involved: S1. Acquire historical earth-rock dam breach characteristic data and corresponding earth-rock dam breach peak flow, and construct an earth-rock dam breach dataset using the historical earth-rock dam breach characteristic data and corresponding earth-rock dam breach peak flow; S2. Constructing a BP neural network model: dividing the historical earth-rock dam burst data set into a training set and a test set in a ratio of 8:2, constructing the historical earth-rock dam burst feature data into a feature vector P, wherein each historical earth-rock dam burst feature data and its corresponding earth-rock dam burst peak flow are taken as a sample; using the newff function to construct a BP neural network model, wherein the input layer is the feature vector P, and the output layer is the corresponding earth-rock dam burst peak flow; during the network training process, the input data is passed to the hidden layer through the associated weights and biases, and after being processed by the activation function, the output layer is used to give the predicted earth-rock dam burst peak flow; S3. Train and optimize the BP neural network model through the historical earth-rock dam breach data set and genetic algorithm: train the BP neural network model through the train function, and use the back propagation algorithm to optimize the weights and biases of the network during the training process to minimize the mean square error until the preset cutoff condition is reached; then, optimize the weights and biases of the BP neural network model through the genetic algorithm, and assign the weights and biases optimized by the genetic algorithm to the BP neural network model; The method of optimizing the weights and biases of the BP neural network model by genetic algorithm is as follows: Initialize the population, where each individual represents the weight and bias of a network parameter of the BP neural network model, and a uniform distribution is used for individual initialization; Define the fitness function, its formula is: ； In the formula, is the fitness value, is the number of samples in the training set, is the serial number of each sample in the training set, is the predicted peak flow of earth-rock dam breach, is the peak flow of earth-rock dam burst corresponding to the samples in the training set; individuals with higher fitness values ​​are selected to enter the next generation population through the normalized geometric selection method, and the best individuals are retained in each generation by the elite strategy; arithmetic crossover is selected as the crossover method, the genes of the two parent individuals are combined to generate new offspring individuals, and the crossover operation uses the weights and biases of the two parents to reorganize and generate new individuals; then the weights and biases of the individuals are mutated by the non-uniform mutation method, and the mutation includes small-scale random changes; then, in each iteration of the mutation operation, if the fitness value of the mutated individual is the same as that of the current best individual, the individual is kept as the optimal solution until the preset iteration conditions are met and the optimal solution is found, and this optimal solution is used as the weight and bias of the BP neural network model; The crossover operation formula is: ； ； In the formula, is the cross coefficient, and are the genes of the two parent individuals, and They are two offspring individuals generated after the crossover of two parent individuals; The operational formula of the mutation is: ； In the formula, For the individual after mutation, is the individual before mutation, δ is the randomly generated mutation amount; Afterwards, the trained GA-BP neural network model based on the elite strategy improvement is obtained; S4. Obtain the current earth-rock dam burst characteristic data, input the current earth-rock dam burst characteristic data into the trained GA-BP neural network model based on the elite strategy improvement, and obtain the predicted earth-rock dam burst peak flow.

2. The method for predicting peak flow of earth-rock dam breach based on elite strategy according to claim 1 is characterized in that: The historical earth-rock dam failure characteristic data include dam height, dam top width, dam bottom width, dam width, initial water depth, median particle size, whether grooves are opened, reservoir capacity and porosity.

3. The method for predicting peak flow of earth-rock dam breach based on elite strategy according to claim 2 is characterized in that: Before constructing the BP neural network model, the data in the historical earth-rock dam failure characteristic data are normalized in turn.

4. The method for predicting peak flow of earth-rock dam breach based on elite strategy according to claim 1 is characterized in that: In step S4, the method of inputting the current earth-rock dam breach characteristic data into the trained GA-BP neural network model based on the elite strategy improvement to obtain the predicted earth-rock dam breach peak flow is: Collect current earth-rock dam breach characteristic data and normalize the current earth-rock dam breach characteristic data; The normalized current earth-rock dam breach characteristic data is input into the trained GA-BP neural network model based on the elite strategy improvement; Furthermore, the trained GA-BP neural network model based on the elite strategy improvement can output the predicted peak flow of earth-rock dam breach.

5. The method for predicting peak flow of earth-rock dam breach based on elite strategy according to claim 1 is characterized in that: The specific process of transferring the historical earth-rock dam failure characteristic data in the input layer to the hidden layer through the associated weights and biases, and the output layer giving the predicted earth-rock dam failure peak flow is as follows: Select the logsig formula as the activation function. The activation function is as follows: ； In the formula, is the input value, is the base of natural logarithms; The data passed to the hidden layer is the weighted and biased data of the input layer. Assume that the input layer has n nodes and the hidden layer has m nodes. The input data is , , the weight is , the bias is ,in, is the weight from the jth input layer node to the ith hidden layer node, is the weight from the jth input layer node to the i+1th hidden layer node, is the bias of the i-th hidden layer node; then, the input of the hidden layer node is the weighted sum of the i-th hidden layer node, expressed as the following formula: ； Finally, the output layer is calculated based on the output of the hidden layer to generate the predicted earth-rock dam breach peak flow rate, which is obtained by the following formula: ； In the formula, is the weight from the mth hidden layer node to the output layer, is the predicted peak flow of earth-rock dam failure, represents the bias of the jth node in the hidden layer, is the bias of the output layer.

Images