Intelligent prediction method for dam seepage pressure considering environmental load hysteresis

Abstract

The application discloses a dam seepage pressure explainable intelligent prediction method considering environmental load hysteresis, relates to the technical field of dam operation safety monitoring and management, and comprises the following steps: using time hysteresis effect, mining an environmental load factor set, and establishing a dam seepage pressure causal analysis mathematical model based on the environmental load factor set; further, on the basis of extracting deep features of monitoring data by using a variational autoencoder optimized by an attention mechanism, mining a nonlinear function relationship between the dam seepage pressure and influencing factors by using a bidirectional long short-term memory network optimized by a Bayesian optimization method, constructing a deep intelligent prediction model of the dam seepage pressure considering the load hysteresis effect, and introducing a SHAP additive explanation algorithm to quantitatively interpret the contribution degree of input features in the constructed deep intelligent prediction model to a prediction result. The application provides a feasible idea for quantitatively analyzing the influence of dam seepage pressure change under the action of environmental load and high-precision prediction of future change trends of the dam seepage pressure.

Description

Technical Field

[0001] This invention relates to the field of dam operation safety monitoring and management technology, and in particular to an interpretable intelligent prediction method for dam seepage pressure considering environmental load lag. Background Technology

[0002] Seepage analysis is a crucial tool for assessing the "health status" of a dam and a key technology for understanding its seepage resistance under varying hydraulic loads, providing a theoretical framework for dam safety. To strengthen safety measures, it is essential to translate the analytical results into practical prevention and control methods. Seepage monitoring serves as a bridge between theory and practice, not only verifying the analytical results but also providing timely feedback on the dam's condition through data and enabling prediction of seepage during dam operation. Traditional methods often rely on experience or fixed time intervals with lag days, and suffer from poor nonlinear modeling capabilities, simplistic uncertainty handling, and poor interpretability.

[0003] Currently, seepage monitoring mainly falls into three categories: traditional physical methods, machine learning-based research and applications, and others. Physical methods often employ piezometers, manifolds, distributed optical fibers, resistivity imaging, and ground-penetrating radar. Thirdly, based on modeling techniques, different mathematical models can be used to address the same problem. Finite difference methods (FDM), finite element methods (FEM), finite volume methods (FVM), boundary element methods (BEM), and meshless methods are increasingly popular among scientists and engineers. While these physical methods offer high accuracy, they are time-consuming and labor-intensive, making them unsuitable for long-term monitoring. With the development of artificial intelligence, machine learning-based applications have become a more efficient and labor-saving approach, yielding numerous research results.

[0004] In the process of predicting dam seepage pressure, there is a close coupling relationship between two important components: the environmental lag effect and the machine learning model. The appropriate determination of the lag term not only affects the scientific validity and engineering application value of the seepage pressure prediction but also directly determines the accuracy of the prediction. Simultaneously, selecting an appropriate machine learning model plays a crucial role in extracting effective features from complex input-output relationships and establishing high-precision seepage pressure prediction.

[0005] To address these issues, there is an urgent need to consider interpretable intelligent prediction methods for dam seepage pressure with environmental load lag. Summary of the Invention

[0006] To address the aforementioned issues, this application proposes an interpretable intelligent prediction method for dam seepage pressure that considers environmental load lags. The aim is to overcome the shortcomings of existing technologies. This method employs delayed mutual information to quantify the time-lag effects of environmental factors such as reservoir water level and rainfall, constructing a set of environmental load factors influencing dam seepage pressure changes. Based on this, a causal mathematical model for dam seepage pressure analysis is established. Furthermore, based on extracting deep features from monitoring data using an attention-optimized variational autoencoder (VAE), a Bayesian algorithm-optimized bidirectional long short-term memory network (BiLSTM) is used to mine the nonlinear functional relationship between dam seepage pressure and its influencing factors. Thus, a deep intelligent prediction model for dam seepage pressure based on Attention-VAE-BiLSTM is proposed. The SHAP algorithm is then used to quantitatively interpret the established prediction model, quantitatively analyzing the impact of different environmental load factors on dam seepage pressure changes. This invention provides a feasible approach for quantitatively analyzing the impact of environmental loads on dam seepage pressure changes and accurately predicting its future trends.

[0007] This invention provides an interpretable intelligent prediction method for dam seepage pressure considering environmental load lag, comprising the following steps:

[0008] S1. Analyze the engineering monitoring data to obtain initial data, including upstream water level, rainfall, downstream water level and seepage pressure measurements at various monitoring points of the dam. Based on the basic principles of the dam seepage pressure monitoring model, determine the environmental load and quantify the time lag effect of environmental factors to construct a set of environmental load factors that affect the change of dam seepage pressure.

[0009] S2. Divide the dam seepage pressure and environmental load factor set into training set and test set according to a certain ratio. Use the training set to train the parameters of the deep intelligent model. At the same time, use the SHAP method to analyze the contribution of the input features in the deep intelligent prediction model to the prediction results, and then obtain the interpretable intelligent prediction model of dam seepage pressure considering the lag of environmental load.

[0010] S3. The interpretable intelligent prediction results of dam seepage pressure are obtained by using the dam seepage pressure interpretable intelligent prediction model that takes into account the lag of environmental load.

[0011] Preferably, in S1, initial data is obtained by analyzing engineering monitoring data. Based on the basic principles of the dam seepage pressure monitoring model, environmental loads are determined, and the time lag effect of environmental factors is quantitatively analyzed. The specific content of constructing the environmental load factor set affecting changes in dam seepage pressure includes:

[0012] The environmental factors include reservoir water level, rainfall, and downstream water level. The specific steps are as follows:

[0013] S11. The time lag effect of reservoir water level and rainfall on seepage pressure is quantified by the delayed mutual information method to determine the lag days of reservoir water level and rainfall;

[0014] S12. Construct a causal mathematical model for dam seepage pressure that includes reservoir water level, rainfall and its lag factors and time-dependent factors. Determine the influence of reservoir water level and rainfall by fitting the influence duration and verify the physical rationality to obtain the set of environmental load factors affecting the change of dam seepage pressure.

[0015] Preferably, seepage pressure monitoring is the core of dam safety monitoring, reflecting the intensity and stability of seepage within the dam body and foundation, which directly relates to the structural stability and long-term operational safety of the dam. Extensive seepage analysis and engineering practice in earth-rock dams show that dam seepage pressure is mainly determined by the hydrological environment (such as upstream water level, rainfall, and downstream water level) and structural characteristics (such as compaction degree and structural form). Among these, environmental factors are the primary drivers of changes in dam seepage behavior. Based on the fundamental theory of dam safety monitoring, the basic expression of the mathematical analysis model for dam seepage pressure is:

[0016] ;

[0017] in: Indicates the upstream water level component; Indicates the downstream water level component; Indicates the component of rainfall; The failure component is represented by h; h is the water level in the piezometer.

[0018] Preferably, the expression for determining the lag days of reservoir water level and rainfall on seepage pressure by quantifying the time lag effect of reservoir water level and rainfall using the delayed mutual information method is as follows:

[0019] ;

[0020] ;

[0021] ;

[0022] in, For the entropy of environmental variables, The range of values ​​a variable can take. Let x be the probability density function of the variable x. For the environment variable matrix, The environmental load factor at a certain moment. For the target penetration pressure, This is the osmotic pressure matrix. The osmotic pressure value at a certain moment. This represents the mutual information value between environmental variables and osmotic pressure. Let be the joint probability density function. This is the joint probability density function of environmental variables and osmotic pressure;

[0023] Define environmental load sequence There are n data points in total. X As a source of information;

[0024] The information source has a delay time of t, at which point the information receiving point... Y There are a total of nt data points, and ;

[0025] When t is X and Y When considering the time delay between x1 and y, it is assumed that x1 and y 1+t The relationship is the closest, expressed as: ,in i At a certain point in time, X and Y The Directional Information Transfer Index (DITI) between them is calculated as follows:

[0026] ;

[0027] in, It represents the information transmission capability between X and Y.

[0028] Preferably, in S2, the upstream water level component is usually proportional to the first power of the reservoir water level, and due to the lag process of seepage, its modeling factor should consider the influence of previous reservoir water level changes; the rainfall component, with some rainwater infiltrating the dam body and causing piezometer water level changes, also has a lag process; while the downstream water level component changes less, so the downstream water level of the current day is used as the downstream water level component factor; the time-dependent factor, due to the changes in the particle size distribution of the earth-rock dam soil structure, has a time-dependent process, and is usually represented by the sum of a logarithmic function and a linear function. Therefore, the mathematical model for the causal analysis of dam seepage pressure can be expressed as:

[0029] ;

[0030] In the formula: The regression coefficients for the upstream water level component; This refers to the upstream water level on that day. This refers to the rainfall on that day. To monitor the day before t The average reservoir water level over the days; t Take 1, 3, 5... m 1, m 1 This refers to the number of days the upstream water level lagged behind. The regression coefficients for the rainfall component; This refers to the amount of rainfall in the preceding period; r Take 1, 3, 5...m 2, m 2 This refers to the number of days the rainfall was delayed. The regression coefficient for the downstream water level; To monitor downstream water levels daily; , For the time-dependent component regression coefficient; Divide the number of days since the start of the initial water storage period by 100; This is a constant term.

[0031] Preferably, in S2, the training set of dam seepage pressure and its environmental load factors is used for training. The parameters of the deep intelligent prediction model are optimized by Bayesian method, and then a deep intelligent prediction model of dam seepage pressure considering the load lag effect is constructed. SHAP is used to quantify the contribution of input features to the prediction results.

[0032] The specific content of S2 is as follows:

[0033] S21. Divide the dam seepage pressure and environmental load factor set in chronological order at a ratio of 8:2 to obtain the training set and prediction set, and perform standardization preprocessing on the training set and prediction set.

[0034] S22. Use the standardized preprocessed training set to train the parameters in the deep intelligent prediction model to obtain the parameters of the deep intelligent prediction model.

[0035] S23. Using the deep intelligent prediction model with the above-trained parameters, a test set is used to make predictions and obtain the dam seepage pressure prediction results. The prediction results are then introduced into the SHAP analysis model to interpret and quantify the prediction results, resulting in an interpretable intelligent prediction model for dam seepage pressure that takes into account the lag of environmental loads. The contribution of each environmental load factor is visualized and analyzed to intuitively obtain the relationship between the prediction results and the input features.

[0036] The specific details of obtaining the parameters of the deep intelligent prediction model by training the parameters of the deep intelligent prediction model using the standardized preprocessed training set in S22 are as follows:

[0037] The deep intelligent prediction model outputs a bidirectional hidden state H through a Bayesian-optimized BiLSTM and applies an attention mechanism to this bidirectional hidden state to calculate the context vector c.

[0038] The computed bidirectional hidden state H and context vector c are fused and input into the VAE encoder to obtain the logarithmic variance and mean, and then input into the decoder through reparameterized sampling z.

[0039] The decoder input is a fusion of latent variable z, original features, BiLSTM output at the last time step, and Attention context vector, and outputs the predicted value.

[0040] The Adam optimizer is used during training to avoid overfitting.

[0041] This method is used to explore the nonlinear functional relationship between dam seepage pressure and its influencing factors, until the training is completed and the parameters of the deep intelligent prediction model are obtained.

[0042] In summary, this invention provides an interpretable intelligent prediction method for dam seepage pressure considering the lag effect of environmental loads. Compared with traditional technologies, the present invention offers the following advantages:

[0043] 1. The delayed mutual information method is used to quantify the lag effect of environmental factors such as reservoir water level and rainfall on dam seepage pressure, and a mathematical model for causal analysis of dam seepage pressure is constructed accordingly.

[0044] 2. By optimizing the variational autoencoder with a Bayesian-optimized bidirectional long short-term memory network and attention mechanism to extract deep features from monitoring data and fully exploring the nonlinear functional relationship between dam seepage pressure and its environmental effects for prediction, and then combining the SHAP algorithm to quantitatively interpret the prediction model, we can quantitatively analyze the impact of different environmental load factors on the changes in dam seepage pressure, providing a new approach for interpretable, intelligent and accurate prediction of dam seepage pressure during operation.

[0045] The technical method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0046] Figure 1 Flowchart for constructing the interpretable intelligent prediction method for dam seepage pressure considering environmental load lag in this invention;

[0047] Figure 2 A graph showing the time variation of dam monitoring data;

[0048] Figure 3 This is a diagram showing the impact of the lag effect of upstream water level. Figure 3 In the figure, (a) represents the orientation information index of the upstream water level on the K16A osmotic pressure gauge. Figure 3 (b) in the figure represents the orientation information index of the upstream water level on the K15A osmotic pressure gauge. Figure 3 (c) in the figure represents the orientation information index of the upstream water level on the K14B osmotic pressure gauge. Figure 3 (d) in the figure represents the orientation information index of the upstream water level on the K14A osmotic pressure gauge;

[0049] Figure 4 This is a diagram showing the impact of the lag effect of rainfall. Figure 4 In the figure, (a) represents the orientation information index of rainfall on the K16A osmotic pressure gauge. Figure 4 (b) in the figure represents the orientation information index of rainfall on the K15A osmotic pressure gauge. Figure 4 (c) in the figure represents the orientation information index of rainfall on the K14B osmometer. Figure 4 (d) in the figure represents the orientation information index of rainfall on the K14A osmometer.

[0050] Figure 5 This is a comparison chart of the actual, predicted, and fitted values ​​from the model regression. Figure 5 (a) in the figure is a curve showing the fitted value and the actual value of the K14A osmometer. Figure 5 (b) in the figure is a curve showing the fitted value and the actual value of the K14B osmometer. Figure 5 (c) in the figure is a curve showing the fitted value and the actual value of the K15A osmometer. Figure 5 In the figure, (d) is a curve showing the fitted value and the actual value of the K16A osmotic pressure gauge;

[0051] Figure 6 A graph showing the SHAP values ​​of key influencing factors. Figure 6 (a) in the text is H 16-40 Scatter plot of SHAP values Figure 6 (b) in the figure is a scatter plot of the SHAP values ​​of the downstream water level. Figure 6 (c) in the figure is a scatter plot of the SHAP values ​​of the aging component θ. Figure 6 (d) in the figure is a scatter plot of the SHAP values ​​of the upstream water level;

[0052] Figure 7 For feature analysis of bee colony diagrams;

[0053] Figure 8 For feature analysis of bar charts;

[0054] Figure 9 This invention presents a step-by-step diagram illustrating the interpretable intelligent prediction method for dam seepage pressure considering environmental load lag. Detailed Implementation

[0055] The technical method of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0056] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0057] Techniques, systems, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the instruction manual.

[0058] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0059] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0060] Example 1

[0061] This invention provides an interpretable intelligent prediction method for dam seepage pressure considering environmental load lag, such as... Figure 9 As shown, it includes the following steps:

[0062] S1. Analyze the engineering monitoring data to obtain initial data, which includes upstream water level, rainfall, downstream water level, and seepage pressure measurements at various monitoring points on the dam. Based on the basic principles of the dam seepage pressure monitoring model, determine the environmental load and quantify the time lag effect of environmental factors to construct a set of environmental load factors that affect the change of dam seepage pressure.

[0063] Furthermore, in S1, the engineering monitoring data is analyzed to obtain initial data. Combining the basic principles of the dam seepage pressure monitoring model, the environmental load is determined and the time lag effect of environmental factors is quantitatively analyzed. The specific content of constructing the environmental load factor set that affects the change of dam seepage pressure includes: the environmental factors include reservoir water level, rainfall, and downstream water level. The specific steps are as follows: S11, quantify the time lag effect of reservoir water level and rainfall on seepage pressure through the delayed mutual information method, and determine the duration of the influence of reservoir water level and rainfall lag days.

[0064] S12. Construct a mathematical model for causal analysis of dam seepage pressure that includes lag factors and time-dependent factors of reservoir water level and rainfall. Determine the degree of influence of reservoir water level and rainfall by fitting the influence duration and verify the physical rationality.

[0065] Furthermore, seepage pressure monitoring is the core of dam safety monitoring, reflecting the intensity and stability of seepage within the dam body and foundation, which directly relates to the structural stability and long-term operational safety of the dam. Extensive seepage analysis and engineering practice in earth-rock dams show that dam seepage pressure is primarily determined by the hydrological environment (such as upstream water level, rainfall, and downstream water level) and structural characteristics (such as compaction degree and structural form). Among these, environmental factors are the main drivers of changes in dam seepage behavior. Based on the fundamental theory of dam safety monitoring, the basic expression of the mathematical analysis model for dam seepage pressure is:

[0066] ;

[0067] in: Indicates the upstream water level component; Indicates the downstream water level component; Indicates the component of rainfall; The failure component is represented by h; h is the water level in the piezometer.

[0068] Furthermore, by quantifying the time lag effect of reservoir water level and rainfall on seepage pressure using the delayed mutual information method, the expressions for the lag days of reservoir water level and rainfall are determined as follows:

[0069] .

[0070] .

[0071] .

[0072] in, For the entropy of environmental variables, The range of values ​​a variable can take. Let x be the probability density function of the variable x. For the environment variable matrix, The environmental load factor at a certain moment. For the target penetration pressure, This is the osmotic pressure matrix. The osmotic pressure value at a certain moment. This represents the mutual information value between environmental variables and osmotic pressure. Let be the joint probability density function. This is the joint probability density function of environmental variables and osmotic pressure.

[0073] Define environmental load sequence There are n data points in total. X As a source of information.

[0074] The information source has a delay time of t, at which point the information receiving point... Y There are a total of nt data points, and .

[0075] When t is X and Y When considering the time delay between x1 and y, it is assumed that x1 and y 1+t The relationship is the closest, expressed as: ,in i At a certain point in time, X and Y The Directional Information Transfer Index (DITI) between them is calculated as follows:

[0076] .

[0077] in, It represents the information transmission capability between X and Y.

[0078] Furthermore, in S2, the upstream water level component is usually proportional to the first power of the reservoir water level, and due to the lag process of seepage, its modeling factor should consider the influence of previous reservoir water level changes; the rainfall component, with some rainwater infiltrating the dam body and causing piezometer water level changes, also has a lag process; while the downstream water level component changes less, so the downstream water level of the current day is used as the downstream water level component factor; the time-dependent factor, due to the changes in the particle size distribution of the earth-rock dam soil structure, has a time-dependent process, and is usually represented by the sum of a logarithmic function and a linear function. Therefore, the mathematical model for the causal analysis of dam seepage pressure is expressed as:

[0079] .

[0080] In the formula: The regression coefficients for the upstream water level component; This refers to the upstream water level on that day. This refers to the rainfall on that day. To monitor the day before t The average reservoir water level over the days; t Take 1, 3, 5... m 1, m 1 This refers to the number of days the upstream water level lagged behind. The regression coefficients for the rainfall component; This refers to the amount of rainfall in the preceding period; r Take 1, 3, 5... m 2, m 2 This refers to the number of days the rainfall was delayed. The regression coefficient for the downstream water level; To monitor downstream water levels daily; , For the time-dependent component regression coefficient; Divide the number of days since the start of the initial water storage period by 100; This is a constant term.

[0081] S2. Divide the dam seepage pressure and environmental load factor set into training set and test set according to a certain ratio. Use the training set to train the parameters of the deep intelligent model. At the same time, use the SHAP method to analyze the contribution of the input features in the deep intelligent prediction model to the prediction results, and then obtain the interpretable intelligent prediction model of dam seepage pressure considering the lag of environmental load.

[0082] Furthermore, in S2, the dam seepage pressure and its environmental load factor set are used for training. The parameters of the deep intelligent prediction model are optimized by Bayesian method, and then a deep intelligent prediction model of dam seepage pressure considering the load lag effect is constructed. SHAP is used to quantify the contribution of input features to the prediction results.

[0083] The specific content of S2 is as follows:

[0084] S21. Divide the dam seepage pressure and environmental load factor sets in 8:2 ratio according to time sequence to obtain training set and prediction set, and perform standardization preprocessing on training set and prediction set.

[0085] S22. Use the standardized preprocessed training set to train the parameters in the deep intelligent prediction model to obtain the parameters of the deep intelligent prediction model.

[0086] S23. Using the deep intelligent prediction model with the above-trained parameters, a test set is used to make predictions and obtain the dam seepage pressure prediction results. The prediction results are then introduced into the SHAP analysis model to interpret and quantify the prediction results, resulting in an interpretable intelligent prediction model for dam seepage pressure that takes into account the lag of environmental loads. The contribution of each environmental load factor is visualized and analyzed to intuitively obtain the relationship between the prediction results and the input features.

[0087] The specific content of using the standardized preprocessed training set in S22 to train the parameters of the deep intelligent prediction model is as follows: the deep intelligent prediction model outputs a bidirectional hidden state H through a Bayesian optimized BiLSTM, and applies an attention mechanism to the bidirectional hidden state to calculate the context vector c.

[0088] The computed bidirectional hidden state H and context vector c are fused and input into the VAE encoder to obtain the logarithmic variance and mean, which are then input into the decoder through reparameterized sampling z.

[0089] The decoder input is a fusion of latent variable z, original features, BiLSTM output at the last time step, and Attention context vector, and outputs the predicted value.

[0090] The Adam optimizer is used during training to avoid overfitting.

[0091] This method is used to explore the nonlinear functional relationship between dam seepage pressure and its influencing factors, until the training is completed and the parameters of the deep intelligent prediction model are obtained.

[0092] S3. The interpretable intelligent prediction results of dam seepage pressure are obtained by using the dam seepage pressure interpretable intelligent prediction model that takes into account the lag of environmental load.

[0093] To quantify the fit and accuracy of the prediction model during forecasting, the coefficient of determination R², mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) were used to calculate the statistical indices for the model fit and prediction segment at each measurement point.

[0094] After training a deep intelligent prediction model on the training set, 100 samples were randomly selected from the test set for SHAP interpretability analysis to quantitatively interpret the prediction model. The impact of different environmental load factors on the dam's seepage pressure was quantitatively analyzed, the influence indicators of environmental loads on dam seepage pressure were quantified, and the key influencing factors of dam seepage pressure were identified.

[0095] Example 2

[0096] A flowchart of an interpretable intelligent prediction method for dam seepage pressure considering load hysteresis effects is shown below. Figure 1 As shown in the example, this earth-rock dam is located in an earthquake-prone area of ​​my country. The dam has a maximum height of 29m, a crest length of 260m, a width of 8m, a crest elevation of 1875.82m, a total reservoir capacity of 65 million m³, and a normal water level of 1874.40m. It is a medium-sized reservoir primarily for irrigation, industrial and urban domestic water use, combined with flood control and power generation. The region has a continental arid climate, with annual rainfall ranging from 107.2 to 259.3 mm and 33 to 57 rainy days per year, with uneven rainfall distribution throughout the year.

[0097] A 6.9-magnitude earthquake and a 5.2-magnitude aftershock occurred successively in the area where a certain earth-rock dam is located, with a focal depth of approximately 10 km. The reservoir where the earth-rock dam is located is about 90 km from the epicenter, within the earthquake's impact range. To avoid a decrease in the accuracy of monitoring data after the earthquake, the data period monitored in the period before the earthquake was selected as the data period for this invention. To understand the seepage status of the dam body and foundation, a seepage monitoring system was installed on the dam, with a total of 4 monitoring cross sections. The cross sections were shared by the seepage pressure measuring points of the dam body and foundation. Among them, there are 16 seepage pressure measuring points in the foundation, with 4, 4, 5, and 3 measuring points respectively arranged in the four cross sections; there are 8 seepage pressure measuring points in the dam body, with 2 measuring points arranged in each cross section. The monitoring equipment uses directly buried piezometers to achieve automated data acquisition. This invention selects the 0+275 cross section, which has the best data continuity and integrity, as the object. This cross section has 3 seepage pressure measuring points in the foundation and 2 seepage pressure measuring points in the dam body. Figure 2 As shown. The monitoring data series covers the dynamics of seepage pressure at each monitoring point, upstream and downstream water level processes, and basin rainfall data. The characteristics of changes in each variable during typical periods are shown in the figure. Figure 3 As shown.

[0098] The process is as follows:

[0099] Step 11. Quantify the time lag effect of reservoir water level and rainfall on seepage pressure using the delayed mutual information method, determine the significant impact duration of reservoir water level and rainfall lag days, and ensure that the input characteristics conform to the actual physical delay law;

[0100] ;

[0101] ;

[0102] ;

[0103] Assuming environmental load sequence There are n data points, and X acts as the information source. If the information source delay time is t, there are nt information receiving points Y. When t is the delay time between X and Y, it can be considered to some extent that x1 and y... 1+t The relationship is the closest, and the mathematical expression is: .

[0104] ;

[0105] The application of DITI quantifies the relationship between X and Y and describes the coupling relationship between them; as shown in Table 1, the influence of rainfall lag effect is illustrated in the figure. Figure 4 As shown;

[0106] Table 1. Number of Lag Days

[0107] ;

[0108] Step 12. Construct a mathematical model for causal analysis of dam seepage pressure that includes the lag days of reservoir water level and rainfall, as well as the time factor. Determine the degree of influence of reservoir water level and rainfall by fitting historical data and verify the physical rationality.

[0109] Step 13. The factors affecting seepage in earth-rock dams include upstream water level, rainfall, time-dependent components, and downstream water level, among which upstream water level and rainfall have the most significant impact.

[0110] ;

[0111] in: Indicates the upstream water level component; Indicates the downstream water level component; Indicates the component of rainfall; The failure component is represented by h; h is the water level in the piezometer.

[0112] S14. From this, we can obtain the mathematical model for the causal analysis of dam seepage pressure:

[0113] ;

[0114] In the formula: The regression coefficients for the upstream water level component; This refers to the upstream water level on that day. This refers to the rainfall on that day. To monitor the day before t The average reservoir water level over the days; t Take 1, 3, 5... m 1, m 1 This refers to the number of days the upstream water level lagged behind. The regression coefficients for the rainfall component; This refers to the amount of rainfall in the preceding period; r Take 1, 3, 5... m 2, m 2 This refers to the number of days the rainfall was delayed. The regression coefficient for the downstream water level; To monitor downstream water levels daily; , For the time-dependent component regression coefficient; Divide the number of days since the start of the initial water storage period by 100; This is a constant term.

[0115] Step 21. Construct an intelligent prediction model for dam seepage pressure depth considering lag effects by inputting lag variable parameters. In the equation, the average upstream water level and average rainfall are used as regression factors to represent the lag effect, including H1, which is a lag factor of 1 day for the upstream water level; H 6-15 The lag factor is 6-15 days for the upstream water level lag; H 16-40 P1 is the lag factor for upstream water level lag of 16-40 days; P2 is the lag factor for rainfall lag of 1 day; P 6-15 The lag factor is 6-15 days for the lag in rainfall; P 16-40 The lag factor is 16-40 days for the lag of rainfall; Downwater is the lag factor for the actual downstream water level; Upwater is the lag factor for the upstream water level; Rainfall is the lag factor for the actual rainfall; θ and lnθ are two lag factors for the time-dependent components.

[0116] Step 22. Divide the dam seepage pressure and its load factor set in chronological order at an 8:2 ratio to obtain the training set and the prediction set, and perform standardization preprocessing on the training set and the prediction set;

[0117] The standardized preprocessed training set is input into the initial deep intelligent prediction model for training. A bidirectional hidden state H is output through a Bayesian-optimized BiLSTM, and an attention mechanism is applied to this bidirectional hidden state to calculate the context vector c. The calculated bidirectional hidden state and context vector c are fused and input into the VAE encoder to obtain the logarithmic variance and mean, which are then input into the decoder through reparameterized sampling z. The decoder input is a fusion of latent variable z + original features + BiLSTM's last time step output + attention context vector, and outputs the predicted value. An Adam optimizer is used during training to avoid overfitting. This method is used to explore the nonlinear functional relationship between dam seepage pressure and its influencing factors until training is complete, resulting in a deep intelligent prediction model.

[0118] Step 23. Using the deep intelligent prediction model with the trained parameters, perform predictions on the test set to obtain the dam seepage pressure prediction results. Then, introduce the prediction results into the SHAP analysis model for interpretation and quantitative analysis. Through visualization analysis of the contribution of each environmental load factor, intuitively obtain the relationship between the prediction results and the input features, such as... Figure 5 As shown.

[0119] Step 31. During prediction, to quantitatively evaluate the fit and accuracy of the prediction model, the coefficient of determination R², mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) were used to calculate the statistical indicators of the prediction model fit and prediction segment for each measurement point, as shown in Table 2. The fit comparison... Figure 6 As shown.

[0120] Table 2 Model Evaluation Indicators

[0121] ;

[0122] Step 32. After training the deep intelligent prediction model on the training set, randomly select 100 samples from the test set for SHAP interpretability analysis to quantitatively interpret the prediction model. Quantitatively analyze the impact of different environmental load factors on the dam's seepage pressure, quantify the impact indicators of environmental loads on dam seepage pressure, and identify important influencing factors of dam seepage pressure, such as... Figure 7 and Figure 8 As shown.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical methods of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical methods of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical methods to deviate from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for interpretable intelligent prediction of dam seepage pressure considering environmental load lag, characterized in that, Includes the following steps: S1. Analyze the engineering monitoring data to obtain initial data, combine the basic principles of the dam seepage pressure monitoring model, determine the environmental load and quantify the time lag effect of environmental factors, and construct a set of environmental load factors that affect the change of dam seepage pressure. The initial data includes upstream water level, rainfall, downstream water level, and seepage pressure measurements at various monitoring points on the dam. S2. Divide the dam seepage pressure and environmental load factor set into training set and test set. Use the training set to train the parameters of the deep intelligent model. At the same time, use the Shapley additive interpretation algorithm (SHAP) to analyze the contribution of the input features to the prediction results in the deep intelligent prediction model. In this way, we can obtain an interpretable intelligent prediction model for dam seepage pressure that takes into account the lag of environmental load. S3. The interpretable intelligent prediction results of dam seepage pressure are obtained by using the dam seepage pressure interpretable intelligent prediction model that takes into account the lag of environmental load. S1 analyzes engineering monitoring data to obtain initial data. Based on the fundamental principles of the dam seepage pressure monitoring model, it determines environmental loads and quantifies the time lag effect of environmental factors. The specific content of constructing the environmental load factor set affecting dam seepage pressure changes includes: The environmental factors include reservoir water level, rainfall, and downstream water level. The specific steps are as follows: S11. The time lag effect of reservoir water level and rainfall on seepage pressure is quantified by the delayed mutual information method to determine the lag days of reservoir water level and rainfall. S12. Construct a mathematical model for causal analysis of dam seepage pressure, which includes reservoir water level, rainfall and its lag factors and time factors. Determine the influence of reservoir water level and rainfall by fitting the influence duration and verify the physical rationality to obtain the set of environmental load factors affecting the change of dam seepage pressure. In S2, the training set of dam seepage pressure and its environmental load factors is used for training. The parameters of the deep intelligent prediction model are optimized by Bayesian method, and then a deep intelligent prediction model of dam seepage pressure considering the load lag effect is constructed. SHAP is used to quantify the contribution of input features to the prediction results. The specific details of obtaining the parameters of the deep intelligent prediction model by training the parameters of the deep intelligent prediction model using the standardized preprocessed training set in S22 are as follows: The deep intelligent prediction model outputs a bidirectional hidden state H through a Bayesian-optimized BiLSTM and applies an attention mechanism to this bidirectional hidden state to calculate the context vector c. The computed bidirectional hidden state H and context vector c are fused and input into the VAE encoder to obtain the logarithmic variance and mean, and then input into the decoder through reparameterized sampling z. The decoder input is a fusion of latent variable z, original features, BiLSTM output at the last time step, and Attention context vector, and outputs the predicted value. The Adam optimizer is used during training to avoid overfitting.

2. The intelligent prediction method for interpretable seepage pressure of dams considering environmental load lag as described in claim 1, characterized in that, The time lag effect of reservoir water level and rainfall on seepage pressure is quantified using the delayed mutual information method, and the expressions for the lag days of reservoir water level and rainfall are determined as follows: ； ； ； in, For the entropy of environmental variables, The range of values ​​a variable can take. For variables The probability density function, For the environment variable matrix, x The environmental load factor at a certain moment. For the target penetration pressure, This is the osmotic pressure matrix. The osmotic pressure value at a certain moment. This represents the mutual information value between environmental variables and osmotic pressure. Let be the joint probability density function. This is the joint probability density function of environmental variables and osmotic pressure.

3. The intelligent prediction method for interpretable seepage pressure of dams considering environmental load lag as described in claim 2, characterized in that, Define environmental load sequence There are n data points in total. X As a source of information; The information source has a delay time of t, at which point the information receiving point... Y There are a total of nt data points, and ; When t is X and Y When considering the time delay between x1 and y, it is assumed that x1 and y 1+t The relationship is the closest, expressed as: ,in i At a certain point in time, X and Y The Directional Information Transfer Index (DITI) between them is calculated as follows: ； in, It represents the information transmission capability between X and Y.

4. The intelligent prediction method for interpretable seepage pressure of dams considering environmental load lag as described in claim 3, characterized in that, The mathematical model for causal analysis of dam seepage pressure, which includes reservoir water level, rainfall, and their lag and time-dependent factors, is expressed as follows: ； In the formula: The regression coefficients for the upstream water level component; This refers to the upstream water level on that day. This refers to the rainfall on that day; To monitor the day before The average reservoir water level over the days; Take 1, 3, 5, ... , This refers to the number of days the upstream water level lagged behind. The regression coefficients for the rainfall component; This refers to the amount of rainfall in the preceding period; Take 1, 3, 5, ... , This refers to the number of days the rainfall was delayed. The regression coefficient for the downstream water level; To monitor the downstream water level on a daily basis; , For the time-dependent component regression coefficient; Divide the number of days since the start of the initial water storage period by 100; This is a constant term.

5. The intelligent prediction method for interpretable seepage pressure of dams considering environmental load lag as described in claim 4, characterized in that, The specific content of S2 is as follows: S21. Divide the dam seepage pressure and environmental load factor set in chronological order at a ratio of 8:2 to obtain the training set and prediction set, and perform standardization preprocessing on the training set and prediction set. S22. Use the standardized preprocessed training set to train the parameters in the deep intelligent prediction model to obtain the parameters of the deep intelligent prediction model. S23. Using the deep intelligent prediction model with the above-trained parameters, a test set is used to make predictions and obtain the dam seepage pressure prediction results. The prediction results are then introduced into the SHAP analysis model to interpret and quantify the prediction results, resulting in an interpretable intelligent prediction model for dam seepage pressure that takes into account the lag of environmental loads.

Images