Multi-model joint decision-making flood season dam project stability prediction method

Abstract

The invention discloses a flood season dam project stability prediction method based on multi-model joint decision making. The method comprises the following steps of obtaining parameters of a dam to be predicted and setting a judgment index, establishing a numerical model according to the parameters of the dam to be predicted, assigning values to the parameters of the numerical model according to the judgment index, and solving a safety coefficient of the numerical model of the dam. Three single machine learning models are established, an improved grey wolf algorithm is introduced for optimization, and an optimal model is obtained through judgment indexes and corresponding safety coefficient training models; performing decision-level fusion on the optimal model through a D-S evidence theory to construct a prediction model; and solving the safety coefficient of the to-be-predicted dam through the prediction model, and comparing the safety coefficient with the allowable safety coefficient of the to-be-predicted dam to complete dam stability prediction. According to the method, the multiple single machine learning models are optimized and fused to construct the prediction model, the advantages of different models can be effectively fused, and the precision and reliability of dam stability prediction are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dam engineering stability, and in particular to a multi-model joint decision-making method for predicting dam engineering stability during flood season. Background Art

[0002] Dams are a type of water conservancy project built for the purpose of flood control, water conservancy management and water supply. Most dam projects have been built for a long time and have been in service for a long time. The geological conditions are complex and the material properties of the dam are highly uncertain. Under the continuous scouring of turbulent water during the flood season and the infiltration of high flood levels, dams are very prone to cracking, collapse and landslides. Therefore, it is of great engineering significance to study the stability prediction and monitoring methods of dam projects during the flood season.

[0003] At present, the monitoring and prediction of dam project stability usually involves burying sensors inside the dam or arranging monitoring probes on the dam. Although these methods have achieved some results in practical application, burying sensors inside the dam will inevitably reduce the stability of the dam structure, and when monitoring through equipment, the monitoring effect and reliability are seriously affected by the surrounding environment. In addition, many numerical methods are introduced to solve dam stability, such as finite element method, discrete element method and fast Lagrangian difference method, but they have the disadvantage of low efficiency when processing a large number of tasks. With the rapid development of computer technology, machine learning has become an important artificial intelligence technology, and has achieved fruitful research results in prediction research in many fields, opening up new ideas for dam stability prediction. However, different types of machine learning algorithms have their own advantages. When performing complex prediction work, a single type of algorithm model often cannot meet all working conditions. Therefore, considering the fusion application of multiple existing intelligent machine learning algorithms, developing a more systematic and efficient slope stability monitoring and prediction method has become a key issue to be solved in this field. Summary of the invention

[0004] In view of the deficiencies of the prior art, the object of the present invention is to provide a method for predicting the stability of dam projects during flood season with small impact on engineering structures, fast convergence speed and high prediction accuracy, so as to better ensure the safe operation of dam projects during flood season.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for predicting the stability of dam projects during flood season based on multi-model joint decision-making, comprising the following steps: Step S1: Obtain the cross-sectional geometric parameters of the dam to be predicted, the basic physical and mechanical parameters of the rock and soil inside the dam to be predicted, and the peak water level of the dam area during the flood season. and maximum rainfall intensity , taking the basic physical and mechanical parameters of rock and soil, flood peak water level and maximum rainfall intensity as judgment indicators, and performing data amplification; Step S2: According to the cross-sectional geometric parameters of the dam to be predicted, the dam numerical model is established using engineering simulation software. According to the judgment index after data amplification, the parameters of the dam numerical model are assigned and boundary conditions are set. The limit equilibrium method is selected to solve the safety factor of the dam numerical model after the assignment and boundary conditions are set. ; Step S3: constructing the judgment index after data amplification and the corresponding safety factor K into a data set; Step S4: Establish three single machine learning models including Bayesian Ridge Regression Model BRR, Least Squares Support Vector Regression Model LSSVR and Extreme Gradient Boosting Model XGBoost, introduce the improved grey wolf algorithm IGWO to optimize the hyperparameters of the three single machine learning models, and train the three single machine learning models in the data set to obtain the optimal three single machine learning models; Step S5: Using DS evidence theory as the prediction framework, the three best single machine learning models are fused at the decision level to construct a dam safety factor prediction model for multi-model joint decision-making; Step S6: Determine the allowable safety factor of the dam to be predicted The basic physical and mechanical parameters of the rock and soil inside the dam to be predicted, the flood peak water level and the maximum rainfall intensity are input into the dam safety factor prediction model of multi-model joint decision-making to obtain the safety factor of the dam to be predicted output by the prediction model ,Will and For comparison, , then the dam to be predicted is determined to be stable. , then the dam to be predicted is determined to be unstable.

[0006] Furthermore, the specific process of introducing the improved gray wolf algorithm to optimize the hyperparameters of three single machine learning models is as follows: Step S4.21: Select hyperparameters; Choosing the regularization parameter for the Bayesian Ridge regression model and noise variance , the regularization parameter of the least squares support vector regression model and slack variables and the regularization parameter of the extreme gradient boosting model and the number of decision trees As a hyperparameter; Step S4.22: population generation; Initialize the gray wolf population based on the Tent chaotic map, expressed as: ; In the formula, For the The location of the gray wolf; For the The location of the gray wolf; express A random value in the interval; Step S4.23: gray wolf division; Calculate the fitness value of each gray wolf in the initial population, and sort them in order from large to small, and divide the top three gray wolves into Wolf, Wolf and Wolf, the remaining gray wolf as Wolf; Wolf, Wolf, Wolf and The wolves represent the optimal solution, suboptimal solution, general solution, and candidate solution; Step S4.24: simulate the strategy of hunting prey, and introduce the Levy distribution strategy and spiral update method for improvement, update the positions of α wolf, β wolf and δ wolf, expressed as: ; ; ; ; ; In the formula, , and Respectively At iteration Wolf, Wolf, The location of the wolf; , and Respectively At iteration Wolf, Wolf, The location of the wolf; is the fitness function; is the random number of the moving position; represents a random variable of the Levy distribution strategy, is the exponential parameter of the Levy distribution strategy; is point multiplication; The base of natural logarithm Power; is the logarithmic spiral constant; yes Random numbers in ; yes Random numbers in ; express the relative position of the wolf to its prey; Step S4.25: Simulate the prey hunting strategy and update the positions of wolf α, wolf β and wolf δ, expressed as: ; In the formula, For the The position of the gray wolf at the iteration; , and Respectively At iteration Wolf, Wolf, The location of the wolf; Step S4.26: Repeat steps S4.23 to S4.25 until the maximum number of iterations is reached. The location of the wolf is used as the optimal hyperparameter.

[0007] Furthermore, the specific process of step S5 is as follows: Step S5.1: Import the input items into the three best single machine learning models for prediction, and obtain three prediction results, which are recorded as , and , is the predicted value of the Bayesian Ridge regression model, is the predicted value of the least squares support vector regression model, is the predicted value of the extreme gradient boosting model; Step S5.2: Based on DS evidence theory, , and Transformed into basic probability distribution, assuming that the prediction errors of the three best single machine learning models obey the normal distribution, the confidence distribution, i.e. the basic probability distribution, is calculated according to the probability density of the prediction errors of the three best single machine learning models, expressed as: ; In the formula, Representing the best three single machine learning models Support for the first preset The basic probability distribution of express The prediction error of is a known true value; Step S5.3: According to the evidence theory synthesis rules, the basic probability distributions of the three best single machine learning models are fused to obtain the comprehensive basic probability distribution , Indicates the comprehensive support of the preset; Step S5.4: According to , select the one with the largest The prediction result is taken as the final prediction value , expressed as: ; In the formula, Indicates the maximum The corresponding prediction results.

[0008] Furthermore, the cross-sectional geometric parameters of the dam to be predicted are obtained by conducting on-site measurements of the dam to be predicted and integrating the existing survey results; the basic physical and mechanical parameters of the rock and soil inside the dam are obtained by coring and sampling the rock and soil inside the dam to be predicted and conducting basic tests; the peak water level of the dam area during the flood season is and maximum rainfall intensity It is obtained by consulting hydrological and meteorological data; The cross-sectional geometric parameters of the dam to be predicted include the dam height , Width of dike top , Slope ratio of dam facing water Back slope ratio ; The basic physical and mechanical parameters of the rock and soil inside the dam include gravity , Cohesion , internal friction angle , Permeability coefficient and pore pressure ratio .

[0009] Furthermore, the specific process of step S2 is: Step S2.1: Establish a two-dimensional digital model of the dam according to the cross-sectional geometric parameters of the dam to be predicted, and import the established two-dimensional digital model into the engineering simulation software to generate a dam numerical model; Step S2.2: meshing and parameter assignment of the generated dam numerical model; Step S2.3: The boundary conditions of the dam numerical model are set according to the judgment index after data amplification. The boundary conditions include: the bottom and left and right ends of the dam numerical model are set as impermeable boundaries, that is, zero flow boundaries; the part of the dam slope on the upstream side of the dam numerical model below the flood peak water level is set as a constant head boundary, and the part of the dam slope on the upstream side of the dam numerical model above the flood peak water level, the dam top and the dam slope on the back side are all set as rainfall boundaries; Step S2.4: After setting the boundary conditions, the seepage calculation is first performed on the dam numerical model, and then the stability calculation is performed. The results of the seepage calculation and the stability calculation are coupled to obtain the safety factor K of the dam numerical model.

[0010] Furthermore, the specific process of step S3 is: the judgment index after data amplification and the corresponding safety factor are set as a group, and the data set is obtained after aggregation , , , The input items are the physical and mechanical parameters of the rock and soil mass, the peak water level and the maximum rainfall intensity. is the output item, i.e. the safety factor of the dam numerical model .

[0011] Furthermore, three single machine learning models are trained in the data set to obtain the three best single machine learning models. The specific process is as follows: the data set is normalized to the maximum and minimum and then divided into a training set, a validation set, and a test set in a ratio of 3:1:1. The training set and the validation set are used to train and evaluate the three single machine learning models, respectively, to obtain the three best single machine learning models. The mean absolute percentage error (MAPE), root mean square error (RMSE) and coefficient of determination were used. Complete the performance evaluation of three single machine learning models; when MAPE < 5%, RMSE < 0.05 and >95%, the accuracy of the three single machine learning models after training meets the requirements, otherwise re-training is required.

[0012] Furthermore, the specific process of using the test set to evaluate the performance of the prediction model is as follows: importing the test set into the dam safety factor prediction model of multi-model joint decision-making and the best three single machine learning models, comparing the evaluation results of the dam safety factor prediction model of multi-model joint decision-making and the best three single machine learning models, and when the performance of the dam safety factor prediction model of multi-model joint decision-making is lower than the best three single machine learning models, returning to step S4.

[0013] Compared with the existing technology, the present invention has the following beneficial effects:

[0014] (1) The present invention optimizes the single machine learning model by adopting an improved optimization algorithm, and constructs a prediction model by fusing the optimized single machine learning model with the DS evidence theory. This can effectively integrate the advantages of different models, greatly improve the accuracy and reliability of dam stability prediction, and is suitable for dam stability assessment.

[0015] (2) The present invention optimizes the hyperparameters of a single machine learning model by adopting an improved gray wolf optimization algorithm, thereby improving the prediction accuracy and generalization ability of the single machine learning model, avoiding the drawback of traditional optimization methods that easily fall into local optimality, and increasing the stability and computational efficiency of the single machine learning model.

[0016] (3) The present invention adopts DS evidence theory as the prediction framework and integrates three optimized single machine learning models at the decision level to achieve uncertainty processing and credibility allocation of information between models, enhance the robustness and reliability of model prediction, and further improve the accuracy of dam stability prediction under complex working conditions by integrating the advantages of multiple models.

[0017] (4) Compared with the prior art, the present invention has significantly improved prediction speed, accuracy and applicability, especially under complex and changeable dam working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The figure is a flow chart of the method of the present invention.

[0019] Figure 2 Schematic diagram of the geometric parameters of the cross section of the dam to be predicted.

[0020] Figure 3 Schematic diagram of the dam numerical model.

[0021] Figure 4 A flow chart for constructing a dam safety factor prediction model for multi-model joint decision-making according to the present invention. DETAILED DESCRIPTION

[0022] like Figure 1 As shown, the present invention provides a technical solution: a method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making, comprising the following steps:

[0023] Step S1: Conduct on-site measurement of the dam to be predicted, integrate the existing survey results, and obtain the cross-sectional geometric parameters of the dam to be predicted; take core samples from the rock and soil inside the dam to be predicted and conduct basic tests to obtain the basic physical and mechanical parameters of the rock and soil inside the dam; and obtain the peak water level of the dam area during the flood season by consulting the hydrological and meteorological data. and maximum rainfall intensity The basic physical and mechanical parameters of rock and soil, flood peak water level and maximum rainfall intensity are used as judgment indicators, and data amplification is performed.

[0024] Among them, the cross-sectional geometric parameters of the dam to be predicted include the dam height , Width of dike top , Slope ratio of dam facing water Back slope ratio ; The basic physical and mechanical parameters of the rock and soil inside the dam include gravity , Cohesion , internal friction angle , Permeability coefficient and pore pressure ratio ,like Figure 2 shown.

[0025] The height of the dam , Width of dike top , Slope ratio of dam facing water , back water slope ratio , severe , Cohesion , internal friction angle , Permeability coefficient and pore pressure ratio There are internal and external correlations between them and the dam stability, and they are nonlinear, which helps to improve the reliability of the prediction results; specifically, the severe , Cohesion , internal friction angle , Permeability coefficient and pore pressure ratio It is the inherent manifestation of the physical mechanics of the dam. , Permeability coefficient and pore pressure ratio The larger the value, the worse the stability of the dam, that is, the smaller the safety factor, the greater the probability of the dam being damaged and unstable. and internal friction angle The larger the value, the better the stability of the dam, that is, the greater the safety factor, the smaller the probability of the dam being damaged and unstable. The external environmental factors that affect the stability of the dam during the flood season are mainly the infiltration effect of high flood levels and the scouring effect of extreme rainfall. Therefore, the peak water level during the flood season is and maximum rainfall intensity As an external environmental factor affecting the stability of the dam, the peak water level and maximum rainfall intensity The larger it is, the worse the stability of the dam is, that is, the smaller the safety factor is, and the greater the probability of the dam being damaged and unstable.

[0026] Among them, the basic physical and mechanical parameters of rock and soil mass, flood peak water level and maximum rainfall intensity are taken as judgment indicators, and the specific process of data amplification is as follows: the basic physical and mechanical parameters of rock and soil mass, flood peak water level and maximum rainfall intensity are taken as judgment indicators, the judgment indicators are sorted in order from small to large, and then amplified using linear interpolation function; the amplified judgment indicators are randomly combined to obtain 1000~1500 groups of judgment indicators with different combinations.

[0027] Step S2: According to the cross-sectional geometric parameters of the dam to be predicted, a numerical model of the dam is established using engineering simulation software, such as Figure 3 As shown in the figure, according to the judgment index after data amplification, the parameters of the dam numerical model are assigned and the boundary conditions are set. The limit equilibrium method is selected to solve the safety factor of the dam numerical model after the assignment and boundary conditions are set. .

[0028] Step S2.1: According to the cross-sectional geometric parameters of the dam to be predicted, a two-dimensional digital model of the dam is established, and the established two-dimensional digital model is imported into engineering simulation software to generate a numerical model of the dam; the engineering simulation software may be GeoStudio software, etc.

[0029] Step S2.2: Grid the generated dam numerical model and set the weight of the dam numerical model according to the judgment index after data amplification. , Cohesion , internal friction angle , Permeability coefficient and pore pressure ratio , and the rest of the parameters are set to default; grid division can be performed using the grid division module in GeoStudio software.

[0030] Step S2.3: According to the judgment index after data amplification, the boundary conditions of the dam numerical model are set. The boundary conditions include: the bottom and left and right ends of the dam numerical model are set as impermeable boundaries, that is, zero flow boundaries; the part of the dam slope on the water-facing side of the dam numerical model that is lower than the peak water level is set as a constant head boundary, and the part of the dam slope on the water-facing side of the dam numerical model that is higher than the peak water level, the dam top and the back water side slope are all set as rainfall boundaries.

[0031] Step S2.4: In the engineering simulation software, the seepage calculation is first performed on the dam numerical model after the boundary conditions are set, and then the stability calculation is performed. The safety factor of the dam numerical model can be obtained by coupling the results of the seepage calculation and the stability calculation. ; Seepage calculation can be performed using the seepage analysis module in GeoStudio software, and stability calculation can be performed using the stability analysis module in GeoStudio software.

[0032] Step S3: The judgment index and the corresponding safety factor after data amplification Build a dataset.

[0033] The specific process of step S3 is: the judgment index after data amplification and the corresponding safety factor are set as a group, and the data set is obtained after aggregation , , , The input items are the physical and mechanical parameters of the rock and soil mass, the peak water level and the maximum rainfall intensity. is the output item, i.e. the safety factor of the dam numerical model .

[0034] Step S4: Establish three single machine learning models including Bayesian Ridge Regression Model (BRR), Least Squares Support Vector Regression Model (LSSVR) and Extreme Gradient Boosting Model (XGBoost), introduce the Improved Grey Wolf Algorithm (IGWO) to optimize the hyperparameters of the three single machine learning models, and train the three single machine learning models in the data set to obtain the three optimal single machine learning models.

[0035] The specific process of training three single machine learning models in the data set and obtaining the three optimal single machine learning models is as follows: the data set is normalized to the maximum and minimum and then divided into training set, validation set and test set in a ratio of 3:1:1. The training set and validation set are used to train and evaluate the three single machine learning models respectively to obtain the three optimal single machine learning models.

[0036] The mean absolute percentage error (MAPE), root mean square error (RMSE) and coefficient of determination were used. Complete the performance evaluation of three single machine learning models; if MAPE < 5%, RMSE < 0.05 and >95%, the accuracy of the three single machine learning models after training meets the requirements; if the above conditions are not met, retraining is required.

[0037] Among them, the specific process of establishing three single machine learning models including Bayesian Ridge Regression Model (BRR), Least Squares Support Vector Regression Model (LSSVR) and Extreme Gradient Boosting Model (XGBoost) is as follows: Step S4.11: Construct a Bayesian Ridge regression model, expressed as: (1); In the formula, is the output of the Bayesian Ridge regression model; is the weight vector of the Bayesian Ridge regression model; is Gaussian noise with mean 0 and variance The normal distribution of is the noise variance.

[0038] The Bayesian Ridge regression model constrains the weights by introducing a regularized prior, usually using a zero-mean Gaussian distribution as the prior distribution, expressed as: (2); In the formula, for The prior distribution of is a Gaussian distribution with zero mean, is the regularization parameter of the Bayesian Ridge regression model; Indicates that the mean is 0 and the covariance matrix is Gaussian distribution, is the identity matrix.

[0039] Based on the Bayesian Ridge regression model, the posterior distribution is calculated to obtain the weights and noise variance The closed-form solution of is: (3); In the formula, express The posterior distribution of ; Indicates the mean , the covariance is Gaussian distribution of is the inverse of the noise variance.

[0040] The prediction function based on the Bayesian Ridge regression model is: (4); In the formula, is the predicted value of the Bayesian Ridge regression model.

[0041] Step S4.12: The specific process of constructing the least squares support vector regression model is: The optimization objective function of the least squares support vector regression model is defined as: (5); In the formula, represents the optimization objective function of the least squares support vector regression model; is the weight vector of the least squares support vector regression model; is the bias of the least squares support vector regression model; is the regularization parameter of the least squares support vector regression model; is the number of input items of the least squares support vector regression model; is the slack variable; For the slack variables; for The transpose of .

[0042] The Lagrangian function is introduced to solve the optimization objective function of the least squares support vector regression model, which is expressed as: (6); In the formula, is the Lagrangian function; is the Lagrange multiplier; For the Lagrange multipliers; Indicates Input The kernel map of is the least squares support vector regression model. output items.

[0043] Use the Lagrangian function to , , and Calculate the partial derivative and set it to 0 to obtain the linear prediction function of the least squares support vector regression model: (7); In the formula, is the predicted value of the least squares support vector regression model; is the radial basis kernel function.

[0044] It is expressed as: (8); In the formula, Represents the width of the radial basis kernel function; Represents the exponential function operator.

[0045] Step S4.13: The specific process of constructing the extreme gradient boosting model is: Define the optimization objective function of the extreme gradient boosting model, expressed as: (9); (10); In the formula, represents the optimization objective function of the extreme gradient boosting model; represents the loss function; Indicates Group actual value; represents the extreme gradient boosting model Group prediction value; Indicates A decision tree; represents the number of decision trees; Indicates the number of leaf nodes of the decision tree; represents the weight vector of the extreme gradient boosting model; and are the regularization parameters of the extreme gradient boosting model, Used to control the number of leaf nodes , For control size; express The complexity regularization term of .

[0046] Using the optimization objective function of the second-order Taylor expansion limit gradient boosting model, the approximate expression of the newly added decision tree (incremental tree) is obtained: (11); In the formula, Indicates The optimization objective function of the round-robin extreme gradient boosting model; Indicates the newly added Decision Tree right The predicted value of represents the first-order derivative; represents the second-order derivative; express The complexity regularization term of .

[0047] The process of selecting the split point of the decision tree can be expressed as: (12); In the formula, and Respectively represent the gradient sum of the left subtree and the right subtree before splitting; and Respectively represent the second-order gradient sum of the left subtree and the right subtree before splitting; Indicates gain.

[0048] The weight of each leaf node of the decision tree is obtained by optimizing the objective function of the extreme gradient boosting model: (13); In the formula, Indicates the optimal weight of the current leaf node; Represents the gradient sum of the current leaf node; Represents the second-order gradient sum of the current leaf node.

[0049] Among them, the specific process of introducing the improved gray wolf algorithm to optimize the hyperparameters of three single machine learning models is as follows: Step S4.21: Select hyperparameters; Choosing the regularization parameter for the Bayesian Ridge regression model and noise variance , the regularization parameter of the least squares support vector regression model and slack variables and the regularization parameter of the extreme gradient boosting model and the number of decision trees As hyperparameters; the values ​​of these parameters affect the accuracy and computational efficiency of the algorithm, so they are optimized.

[0050] Step S4.22: population generation; Initialize the gray wolf population based on the Tent chaotic map, expressed as: (14); In the formula, For the The location of the gray wolf; For the The location of the gray wolf; express A random value in the interval.

[0051] Step S4.23: gray wolf division; Calculate the fitness value of each gray wolf in the initial population, and sort them in order from large to small, and divide the top three gray wolves into Wolf, Wolf and Wolf, the remaining gray wolf as Wolf; Wolf, Wolf, Wolf and The wolves represent the optimal solution, suboptimal solution, general solution and candidate solution respectively.

[0052] Step S4.24: simulate the strategy of hunting prey, introduce the Levy distribution strategy and spiral update method for improvement, and update the positions of α wolf, β wolf and δ wolf; When the gray wolf is hunting prey, it constantly updates its own position by locating the prey. In order to avoid premature convergence, the Levy distribution strategy and the spiral update method are integrated to improve the positions of the α wolf and the β wolf in the hunting strategy. The position update method of the α wolf, the β wolf and the δ wolf is expressed as: Wolf: (15); (16); Wolf: (17); Wolf: (18); (19); In the formula, , and Respectively At iteration Wolf, Wolf, The location of the wolf; , and Respectively At iteration Wolf, Wolf, The location of the wolf; is the fitness function; is the random number of the moving position; represents a random variable of the Levy distribution strategy, is the exponential parameter of the Levy distribution strategy; is point multiplication; The base of natural logarithm Power; is the logarithmic spiral constant, which takes the value of 1; yes Random numbers in ; yes Random numbers in ; express The relative position of the wolf and its prey.

[0053] Step S4.25: simulate the prey hunting strategy and update the positions of wolf α, wolf β and wolf δ; After continuously updating its position, the gray wolf begins to attack its prey. The process can be expressed as: (20); In the formula, For the The position of the gray wolf at the iteration; , and Respectively At iteration Wolf, Wolf, Wolf's location.

[0054] Step S4.26: Repeat steps S4.23 to S4.25 until the maximum number of iterations is reached. The location of the wolf is used as the optimal hyperparameter.

[0055] Step S5: Using DS evidence theory as the prediction framework, the three best single machine learning models are fused at the decision level to construct a multi-model joint decision-making dam safety factor prediction model, such as Figure 4 As shown in the figure, the performance of the prediction model is evaluated using the test set.

[0056] Step S5.1: Import the input items into the three best single machine learning models for prediction, and obtain three prediction results, which are recorded as , and .

[0057] Step S5.2: Based on DS evidence theory, , and Transformed into basic probability distribution, assuming that the prediction errors of the three best single machine learning models follow a normal distribution, the confidence distribution, i.e., the basic probability distribution, can be calculated based on the probability density of the prediction errors of the three best single machine learning models, expressed as: (twenty one); In the formula, Representing the best three single machine learning models Support for the first preset The basic probability distribution of express The prediction error of is a known true value.

[0058] Step S5.3: According to the evidence theory synthesis rules, the basic probability distributions of the three best single machine learning models are fused to obtain the comprehensive basic probability distribution , Indicates the overall support of the preset.

[0059] Among them, the basic probability distribution of any two of the three best single machine learning models is and , the basic probability distribution after fusion It is expressed as: (twenty two); In the formula, represents the second preset support; Represents the empty set.

[0060] Step S5.4: According to , select the one with the largest The prediction result is taken as the final prediction value , expressed as: (twenty three); In the formula, Indicates the maximum The corresponding prediction results.

[0061] Based on the above, the three best single machine learning models are integrated through the DS evidence theory to construct a dam safety factor prediction model for multi-model joint decision-making.

[0062] Among them, the specific process of using the test set to evaluate the performance of the prediction model is: importing the test set into the dam safety factor prediction model of multi-model joint decision-making and the best three single machine learning models, comparing the evaluation results of the dam safety factor prediction model of multi-model joint decision-making and the best three single machine learning models, if the performance of the dam safety factor prediction model of multi-model joint decision-making is lower than the best three single machine learning models, return to step S4.

[0063] Step S6: Determine the allowable safety factor of the dam to be predicted The basic physical and mechanical parameters of the rock and soil inside the dam to be predicted, the flood peak water level and the maximum rainfall intensity are input into the dam safety factor prediction model of multi-model joint decision-making to obtain the safety factor of the dam to be predicted output by the prediction model ,Will and For comparison, if , then the dam to be predicted is considered stable. , then the dam to be predicted is considered unstable.

[0064] In this embodiment, the allowable safety factor It is set to 1.3. The anti-sliding stability safety factor of the dam is generally between 1.15 and 1.30. In this embodiment, the maximum safety factor is selected to ensure that the dam is safer and more stable.

[0065] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-model joint decision-making method for predicting the stability of dam projects during flood season, characterized in that: The steps include: Step S1: Obtain the cross-sectional geometric parameters of the dam to be predicted, the basic physical and mechanical parameters of the rock and soil inside the dam to be predicted, and the peak water level of the dam area during the flood season. and maximum rainfall intensity , taking the basic physical and mechanical parameters of rock and soil, flood peak water level and maximum rainfall intensity as judgment indicators, and performing data amplification; Step S2: According to the cross-sectional geometric parameters of the dam to be predicted, the dam numerical model is established using engineering simulation software. According to the judgment index after data amplification, the parameters of the dam numerical model are assigned and boundary conditions are set. The limit equilibrium method is selected to solve the safety factor of the dam numerical model after the assignment and boundary conditions are set. ; Step S3: constructing the judgment index after data amplification and the corresponding safety factor K into a data set; Step S4: Establish three single machine learning models including Bayesian Ridge Regression Model BRR, Least Squares Support Vector Regression Model LSSVR and Extreme Gradient Boosting Model XGBoost, introduce the improved grey wolf algorithm IGWO to optimize the hyperparameters of the three single machine learning models, and train the three single machine learning models in the data set to obtain the optimal three single machine learning models; Step S5: Using DS evidence theory as the prediction framework, the three best single machine learning models are fused at the decision level to construct a dam safety factor prediction model for multi-model joint decision-making; Step S6: Determine the allowable safety factor of the dam to be predicted The basic physical and mechanical parameters of the rock and soil inside the dam to be predicted, the flood peak water level and the maximum rainfall intensity are input into the dam safety factor prediction model of multi-model joint decision-making to obtain the safety factor of the dam to be predicted output by the prediction model ,Will and For comparison, , then the dam to be predicted is determined to be stable. , then the dam to be predicted is determined to be unstable.

2. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 1 is characterized by: The specific process of introducing the improved gray wolf algorithm to optimize the hyperparameters of three single machine learning models is as follows: Step S4.21: Select hyperparameters; Choosing the regularization parameter for the Bayesian Ridge regression model and noise variance , the regularization parameter of the least squares support vector regression model and slack variables and the regularization parameter of the extreme gradient boosting model and the number of decision trees As a hyperparameter; Step S4.22: population generation; Initialize the gray wolf population based on the Tent chaotic map, expressed as: ； In the formula, For the The location of the gray wolf; For the The location of the gray wolf; express A random value in the interval; Step S4.23: gray wolf division; Calculate the fitness value of each gray wolf in the initial population, and sort them in order from large to small, and divide the top three gray wolves into Wolf, Wolf and Wolf, the remaining gray wolf as Wolf; Wolf, Wolf, Wolf and The wolves represent the optimal solution, suboptimal solution, general solution, and candidate solution; Step S4.24: simulate the strategy of hunting prey, and introduce the Levy distribution strategy and spiral update method for improvement, update the positions of α wolf, β wolf and δ wolf, expressed as: ； ； ； ； ； In the formula, , and Respectively At iteration Wolf, Wolf, The location of the wolf; , and Respectively At iteration Wolf, Wolf, The location of the wolf; is the fitness function; is the random number of the moving position; represents a random variable of the Levy distribution strategy, is the exponential parameter of the Levy distribution strategy; is point multiplication; The base of natural logarithm Power; is the logarithmic spiral constant; yes Random numbers in ; yes Random numbers in ; express the relative position of the wolf to its prey; Step S4.25: Simulate the prey hunting strategy and update the positions of wolf α, wolf β and wolf δ, expressed as: ； In the formula, For the The position of the gray wolf at the iteration; , and Respectively At iteration Wolf, Wolf, The location of the wolf; Step S4.26: Repeat steps S4.23 to S4.25 until the maximum number of iterations is reached. The location of the wolf is used as the optimal hyperparameter.

3. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 2 is characterized by: The specific process of step S5 is: Step S5.1: Import the input items into the three best single machine learning models for prediction, and obtain three prediction results, which are recorded as , and , is the predicted value of the Bayesian Ridge regression model, is the predicted value of the least squares support vector regression model, is the predicted value of the extreme gradient boosting model; Step S5.2: Based on DS evidence theory, , and Transformed into basic probability distribution, assuming that the prediction errors of the three best single machine learning models obey the normal distribution, the confidence distribution, i.e. the basic probability distribution, is calculated according to the probability density of the prediction errors of the three best single machine learning models, expressed as: ； In the formula, Representing the best three single machine learning models Support for the first preset The basic probability distribution of express The prediction error of is a known true value; Step S5.3: According to the evidence theory synthesis rules, the basic probability distributions of the three best single machine learning models are fused to obtain the comprehensive basic probability distribution , Indicates the comprehensive support of the preset; Step S5.4: According to , select the one with the largest The prediction result is taken as the final prediction value , expressed as: ； In the formula, Indicates the maximum The corresponding prediction results.

4. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 3 is characterized by: The cross-sectional geometric parameters of the dam to be predicted are obtained by conducting on-site measurements of the dam to be predicted and integrating the existing survey results; the basic physical and mechanical parameters of the rock and soil inside the dam are obtained by coring and sampling the rock and soil inside the dam to be predicted and conducting basic tests; The cross-sectional geometric parameters of the dam to be predicted include the dam height , Width of dike top , Slope ratio of dam facing water Back slope ratio ; The basic physical and mechanical parameters of the rock and soil inside the dam include gravity , Cohesion , internal friction angle , Permeability coefficient and pore pressure ratio .

5. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 4 is characterized by: The specific process of step S2 is: Step S2.1: Establish a two-dimensional digital model of the dam according to the cross-sectional geometric parameters of the dam to be predicted, and import the established two-dimensional digital model into the engineering simulation software to generate a dam numerical model; Step S2.2: meshing and parameter assignment of the generated dam numerical model; Step S2.3: The boundary conditions of the dam numerical model are set according to the judgment index after data amplification. The boundary conditions include: the bottom and left and right ends of the dam numerical model are set as impermeable boundaries, that is, zero flow boundaries; the part of the dam slope on the water-facing side of the dam numerical model below the flood peak water level is set as a constant head boundary, and the part of the dam slope on the water-facing side of the dam numerical model above the flood peak water level, the dam top and the back water side dam slope are all set as rainfall boundaries; Step S2.4: After setting the boundary conditions, the seepage calculation is first performed on the dam numerical model, and then the stability calculation is performed. The results of the seepage calculation and the stability calculation are coupled to obtain the safety factor K of the dam numerical model.

6. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 5 is characterized by: The specific process of step S3 is: the judgment index after data amplification and the corresponding safety factor are set as a group, and the data set is obtained after aggregation , , , The input items are the physical and mechanical parameters of the rock and soil mass, the peak water level and the maximum rainfall intensity. is the output item, i.e. the safety factor of the dam numerical model .

7. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 6 is characterized by: The specific process of training three single machine learning models in the data set and obtaining the three best single machine learning models is as follows: the data set is normalized to the maximum and minimum and then divided into a training set, a validation set, and a test set in a ratio of 3:1:

1. The training set and the validation set are used to train and evaluate the three single machine learning models respectively, and the three best single machine learning models are obtained; The mean absolute percentage error (MAPE), root mean square error (RMSE) and coefficient of determination were used. Complete the performance evaluation of three single machine learning models; when MAPE < 5%, RMSE < 0.05 and >95%, the accuracy of the three single machine learning models after training meets the requirements, otherwise re-training is required.

8. The method for predicting the stability of dam projects during flood season by using multiple models for joint decision-making according to claim 7 is characterized by: The specific process of using the test set to evaluate the performance of the prediction model is as follows: importing the test set into the dam safety factor prediction model of multi-model joint decision-making and the best three single machine learning models, comparing the evaluation results of the dam safety factor prediction model of multi-model joint decision-making and the best three single machine learning models, and when the performance of the dam safety factor prediction model of multi-model joint decision-making is lower than the best three single machine learning models, returning to step S4.

Images